Version 6
User Manual
IDENT
© 2006 BRUKER OPTIK GmbH, Rudolf-Plank-Straße 27, D-76275 Ettlingen, www.brukeroptics.com
All rights reserved. No part of this manual may be reproduced or transmitted in any form or by any
means including printing, photocopying, microfilm, electronic systems etc. without our prior written
permission. Brand names, registered trademarks etc. used in this manual, even if not explicitly marked
as such, are not to be considered unprotected by trademarks law. They are the property of their
respective owner.
The following publication has been worked out with utmost care. However, Bruker Optik GmbH does
not accept any liability for the correctness of the information. Bruker Optik GmbH reserves the right to
make changes to the products described in this manual without notice.
This manual is the original documentation for the OPUS spectroscopic software.
Table of Contents
About OPUS IDENT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1
1
Setting Up an Identity Test Method . . . . . . . . . . . . . . . . . . . . . . . . . . . .3
1.1
1.2
1.3
1.4
1.5
1.6
Loading Existing Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.1.1
Methods created by prior OPUS releases . . . . . . . . . . . . . . . . . . . . . 6
Creating New Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
Setting Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Identity Test Limit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
Validating Library . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.5.1
Validation Report . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
Storing Method Files . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2
Performing an IDENT Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .19
3
IDENT Report . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .21
3.1
4
Cluster Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .25
4.1
4.2
4.3
5
Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
4.1.1
Methods to Calculate Spectral Distances . . . . . . . . . . . . . . . . . . . . 28
4.1.2
Cluster Algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
Performing a Cluster Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3D Files/Filelist . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
Conformity Test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .39
5.1
5.2
6
Identity Test Reports . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.1.1
Standard Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.1.2
Factorization Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
Setting up Conformity Test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
Performing Conformity Test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
IDENT Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .51
6.1
6.2
6.3
6.4
Algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
6.1.1
Standard Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
6.1.2
Factorization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
Factorization Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
6.2.1
Scaling to 1st Range and Normalize to Reprolevel . . . . . . . . . . . . 59
Data Preprocessing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
6.3.1
Vector Normalization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
Determining Threshold Value for Identity Test . . . . . . . . . . . . . . . . . . . . . . 67
iii
6.5
6.6
6.7
7
Identity Test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
Class Test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
Validation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
Reference Section . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .77
7.1
7.2
7.3
7.4
7.5
7.6
7.7
7.8
7.9
7.10
7.11
7.12
7.13
Setup Identity Test Method - Load Method . . . . . . . . . . . . . . . . . . . . . . . . . 77
Setup Identity Test Method - Reference Spectra . . . . . . . . . . . . . . . . . . . . . 78
7.2.1
Sorting Reference Spectra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
7.2.2
Missing Reference Spectra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
7.2.3
Options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
7.2.4
Set Sub Library . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
7.2.5
Assign Classes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
Setup Identity Test Method - Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
7.3.1
Preprocessing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
7.3.2
Regions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
7.3.3
Interactive Region Selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
7.3.4
Clear Selected Regions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
7.3.5
Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
7.3.6
Calculate Thresholds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
Setup Identity Test Method – Threshold . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
7.4.1
Maximum Hit + X*SDev . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
7.4.2
Confidence Level . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
7.4.3
Set . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
7.4.4
Group Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
Setup Identity Test Method – Validate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
7.5.1
Validation Report . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
7.5.2
Print . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
Setup Identity Test Method - Store Method . . . . . . . . . . . . . . . . . . . . . . . . 100
Identity Test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
7.7.1
No Reference Defined . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
Cluster Analysis – Load Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
7.8.1
Load Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
7.8.2
General Information . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
Cluster Analysis – Reference Spectra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
Cluster Analysis – Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
7.10.1 Preprocessing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
7.10.2 Regions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
7.10.3 Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
7.10.4 Calculate Distances . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
Cluster Analysis – Report . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
7.11.1 Score Plot . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
7.11.2 Options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
3D File/Filelist . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
7.12.1 File List . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
Cluster Analysis – Store Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
iv
About OPUS IDENT
This manual consists of two parts. The first part describes how to create a userdefined reference library, and generate an IDENT method. Apart from this the
IDENT analysis and the resulting report files are explained in detail, as well as
the theory of the IDENT software test routines. The cluster analysis and
conformity test are described in a separate chapter.
The reference section refers to all IDENT functions and supports you if you
have questions about the IDENT functionalities, or problems while using
IDENT.
Introduction
OPUS IDENT (in the following referred to as IDENT) is a software package
designed to identify substances by their IR spectra. An unknown spectrum (in
the following called test spectrum) is directly compared to reference spectra of a
library. IDENT identifies those reference spectra which are closest equivalent to
the test spectrum, and determines the deviations between these spectra and the
test spectrum. This allows IDENT to identify unknown substances, e.g.
polymers, and to evaluate the conformity degree of a substance with a reference
standard. The latter application is a typical task found in quality control.
To perform an identity test you first need to have a reference library which you
compare the test spectrum with. If no suitable library exists, you have to
measure a set of reference spectra, i.e. at least one spectrum per substance.
However, it is recommended to measure several batches of the same substance
to enable the program to get more information on possible allowed variations.
Before, the samples required have to be classified as identified using common
reference analytics.
If you have already measured a spectrum which you want to identify, the next
step will be to generate an IDENT method. To perform an identity test you can
select several algorithms and define identification parameters by using the
IDENT software.
The results of an identity test are stored in a report which includes the analysis
result, the method used as well as the parameters defined.
Bruker Optik GmbH
OPUS/IDENT
1
2
OPUS/IDENT
Bruker Optik GmbH
Loading Existing Method
1
Setting Up an Identity Test
Method
This chapter describes step by step how to set up a reference spectra library, and
how to generate an IDENT method. You can perform an analysis using this
IDENT method, and the analysis results will be displayed in the form of a
report.
To perform an identity test the following steps are required:
1) Measuring at least 1 reference spectrum per substance
2) Incorporating the reference spectra into a library (the spectra need to
belong to batches of one single substance which have been identified by
means of conventional analytics)
3) Defining a suitable spectral range for identification
4) Selecting a data preprocessing method
5) Generating an IDENT method
6) Measuring test spectra
7) Analyzing test spectra
This chapter assumes theoretical basic knowledge, and therefore only briefly
outlines the methods involved. The theory of the identity test method is
explained in chapter 6.
If you generate an IDENT method, a main spectra library (IP1) will be created.
All sub-libraries (e.g. IP2, IP3, IP4 etc.) which are related to this reference
library will be stored in the same method file which uses the extension *FAA.
The data structure of the complete library is displayed in a kind of browser
window, and OPUS ensures a common (internal) validation of the entire library.
For further details, see chapter 7.
1.1
Loading Existing Method
Start the IDENT software from the OPUS Evaluate menu. Select the Setup
Identity Test Method command.
Bruker Optik GmbH
OPUS/IDENT
3
Setting Up an Identity Test Method
If you want to load an existing method, click on the Load Method button and
select the respective method.
Figure 1: Setup Identity Test Method - Load Method tab (existing method)
Figure 1 exemplifies the exact description of the main library selected,
including the data structure and the specific parameters in the General
Information of Selected Library group field.
The name of the main library (Ident_Demo) is shown in the blue indicator field
on the upper left side. If you click on one of the sub-libraries in the browser
window, i.e. if you move to a different library level, the indicator field changes
its color and description. In general, you can define as many sub-libraries as you
like. Each library level will have a different color.
•
•
•
•
•
•
IP1:
IP2:
IP3:
IP4:
IP5:
IP6:
Blue
Green
Yellow
Orange
Pink
Magenta
If you defined IP7 as the next library level, the color cycle would start from the
beginning, i.e. IP7 would be blue.
4
OPUS/IDENT
Bruker Optik GmbH
Loading Existing Method
The library level you are currently working with is always displayed on the
upper left side of each property tab, except for the Store Method tab. If you use
a method which has not been properly calculated, an error message pops up and
the library levels not yet re-calculated will be displayed in bold in the browser
window.
Figure 2: Error message on the Load Method tab
To rename the library description, click on the Rename Library button. The
following menu pops up:
Figure 3: Rename reference library
Enter the new name of the current library and confirm it by clicking on the OK
button. The new description is automatically displayed in the browser window
on the right and in the General information... group field. The renaming
procedure applies to each library level.
You can also delete libraries. Note that all sub-libraries assigned to a specific
library level will be deleted as well. Select the particular library in the browser
window and press the DEL key on the PC keyboard. A menu pops up and asks
you whether you want to continue or cancel this deleting operation.
There are two possibilities to print the library structure displayed in the browser
window. Activate the Printer option button to have the library structure printed
on a connected printer. The drop-down list includes the following printing
options:
•
•
•
•
Library tree as shown
Whole library tree
Whole library tree and group statistics
Whole library tree and statistics for each spectrum
Select one of the printing options and click on the Print button. If you activate
the PDF option button, the library structure will be printed as a PDF file.
Bruker Optik GmbH
OPUS/IDENT
5
Setting Up an Identity Test Method
Clicking on the Print button opens the following dialog:
Figure 4: Printing as PDF file
Define the appropriate format, the file name and path of the PDF file. Click on
the
1.1.1
button to open the Define PDF File Name dialog.
Methods created by prior OPUS releases
IDENT methods created by prior OPUS releases, and which also include submethods are automatically converted to OPUS 5.0 if you load them into the
IDENT setup. If one of these sub-methods cannot be found in the path defined,
OPUS tries to search in the path where the reference method (*.FAA) is stored.
If the method cannot be found there either, OPUS writes a remark in the *.log
ASCII-file.
Sub-methods containing spectra which are no member of the main library (IP1)
will not be considered for the set-up of the converted OPUS 5.0 method. If
spectra are assigned to more than one sub-method on the same library level,
they will be considered only once. In both cases the library conversion log file
includes an appropriate remark.
At the end of the conversion a dialog pops-up indicating that the conversion has
been finished, but not completed. Confirm this dialog by clicking on the OK
button.
1.2
Creating New Method
If you want to create a new IDENT method, click on the Reference Spectra tab.
To create a reference library, define the spectra to be used for the library. Click
on the Add Spec. for New Group button to add single spectra to a new product
group. Click on Add Spec. to Sel. Group to add single spectra to a selected
product group.
6
OPUS/IDENT
Bruker Optik GmbH
Creating New Method
Figure 5: Setup Identity Test Method – Reference Spectra tab
In both cases, a dialog box opens where you can select the spectra to be used for
the IDENT method. Having loaded spectra, they will be assigned to the
respective group and are given a consecutive ID number. The Sample Name as
well as the Group Name are automatically read from the file parameters.
To see the list of group-specific spectra, click on the
sign in the first column
(figure 5). Click on the
sign to close the list. You can also remove entries
from the file list. Select one or more entries by clicking on the left mouse button
and holding down the Shift or CTR key. Use the DEL key on the PC keyboard to
delete the entry(ies). Before you can delete a group or spectrum, a menu pops
up and asks you whether you really want to delete the spectra selected.
If you want to create new sub-libraries, click on the Set Sub Libraries button.
These sub-libraries are an integral part of the main IDENT method and are
indicated in green in the Sub Library column (figure 5). The following dialog
opens:
Bruker Optik GmbH
OPUS/IDENT
7
Setting Up an Identity Test Method
Figure 6: Set Sub Libraries
As the spectra of the Carbonhydrates group have already been assigned as a
sub-library they are automatically highlighted (figure 6).
The indicator field shows the defined sub-library IP2 level in green. You can
always set sub-libraries exactly one level below the current library. All
available groups defined for the main method, and which have not yet been
assigned to any sub-library on this level are displayed in the Select groups for...
field. Select the New option from the upper right drop-down list and enter a
unique name for the new sub-library. Then, click on the group(s) you want to
assign. If you define a new sub-library name, OPUS automatically checks
whether the new library name does already exist to avoid double naming. If a
name already exists, a dialog pops up requiring unique naming. Click on the
Assign button.
To delete sub-libraries, select them in the selection field and click on the Delete
Sub Library button. A menu pops up and asks you whether you want to continue
or cancel the deleting operation.
Groups which cannot be separated into further sub-libraries can, however, be
assigned to common classes if class identity is sufficient as analysis result.
8
OPUS/IDENT
Bruker Optik GmbH
Creating New Method
To assign groups to a class click on the Assign Classes button on the Reference
Spectra tab. The following dialog opens:
Figure 7: Assign Classes
The Select groups for... selection field includes only groups which have not yet
been a member of a sub-library or class on this level.
Select the New option from the upper right drop-down list and enter a unique
name for the new class. Then, select the group(s) you want to assign and click
on the Assign button. A menu pops up and asks you whether you want to
continue or cancel the assigning operation.
Click on the Options button on the Reference Spectra tab if you want to change
the path of the original and averaged spectra. The Add selected spectra into one
group check box is checked by default. For further details, see chapter 7.2.3.
Bruker Optik GmbH
OPUS/IDENT
9
Setting Up an Identity Test Method
1.3
Setting Parameters
The quality of an IDENT analysis substantially depends on the data
preprocessing method and spectral regions of each spectrum, which both have
been selected for the samples and IDENT method.
For the main library (IP1) it is recommended to use the Standard method and
define a large spectral region, as the spectral noise will be substantially
smoothed. In case of sub-libraries (IP2, IP3 etc.) it may be better to use the
Factorization method and limit the spectral region, which, of course, causes the
spectral noise to be insufficiently smoothed, but the spectrum shows many
significant details.
Note: If you update library data, new spectra have to be re-calculated on each
library level.
Click on the Parameters tab. This tab defines the spectral regions which have to
be considered for identification. You also have to select the preprocessing
method from the drop-down list as well as the IDENT method (algorithm).
Vector normalization is frequently selected as data preprocessing method.
Sometimes, however, you get even better results if you select First and 2nd
Derivative. For further details, refer to chapter 7. Select Vector Normalization
from the Preprocessing drop-down list.
The Always use lowest IP level check box is only enabled on the first library
level as this is a global setting for the entire library. If you activate the check
box, the IDENT analysis will be performed on the lowest IP level available.
This will also be done even if an IDENTICAL TO result has been achieved at
any higher IP level.
10
OPUS/IDENT
Bruker Optik GmbH
Setting Parameters
Figure 8: Setup Identity Test Method – Parameters tab
If you want to limit the IDENT analysis to certain frequency regions, enter them
into the Regions table. Alternatively, you can also define these regions
interactively.
Bruker Optik GmbH
OPUS/IDENT
11
Setting Up an Identity Test Method
Click on the Interactive Region Selection button. A separate window opens
which displays mean spectra.
Figure 9: Interactive Frequency Range Selection
The gray areas indicate the selected frequency limits, and the white spectral
range is the basis for the subsequent evaluation.
To move spectral regions place the cursor on the respective edge between the
white and gray area, hold down the left mouse button and move the regions. To
delete a spectral region, right click on the white area and select Remove from the
pop-up menu. Click on the OK button to confirm the settings, and the
Parameters tab will be displayed again.
You can also add a new frequency region by a right-click on the left or right
window side. Select the Add Region option from the pop-up menu (figure 10).
12
OPUS/IDENT
Bruker Optik GmbH
Identity Test Limit
Figure 10: Interactive Frequency Range Selection with pop-up menu
Select Standard as method on the Parameter tab and click on the Start
Calculation button to calculate the spectral distances.
1.4
Identity Test Limit
Click on the Threshold tab to have the identity test limits displayed. The
threshold of a reference spectrum is the sum calculated from the maximum
distance listed in the average report, plus the product resulting from the standard
deviation (SDev) and any x factor. For each group the threshold values are listed
in the Threshold column (figure 11).
You can enter any factor into the entry field, 0.25 is set by default. Click on the
Set button to confirm your entry. The factor set is valid for all reference spectra,
and the Threshold column is updated accordingly. You can also print the list.
Click on the Print List button. For further details, see chapter 7.4.
In a similar way you can set the value for the Confidence Level. You can enter
any factor between 95 and 99.9999% into the entry field, 99.99% is set by
default. If you click on the Set button to confirm your entry, the Threshold
column is updated accordingly. The Outlier column displays the number of
spectra which are outside the threshold. If you, e.g., set the confidence level to
95%, then 5% of the total number of spectra will be identified as outliers.
Bruker Optik GmbH
OPUS/IDENT
13
Setting Up an Identity Test Method
Figure 11: Setup Identity Test Method – Threshold tab
1.5
Validating Library
When creating a library, first check whether the IDENT parameters selected for
the IDENT method are optimized for all reference spectra, and whether an
unambiguous assignment of test spectra to a group can be ensured. This is done
by a validation procedure which compares each original spectrum with the
average spectra of all groups. To validate the library, click on the Validate tab.
The following dialog opens:
14
OPUS/IDENT
Bruker Optik GmbH
Validating Library
Figure 12: Setup Identity Test Method - Validate tab
Click on the Validate button to start validation. A dialog pops up and asks you
whether to validate this library and all sub-libraries. For further details, see
chapter 7.5.
1.5.1
Validation Report
The validation result is displayed in the form of a report and stored in a file
which uses the extension *.VAL. Validation reports can also be created for a
single reference library, sub-library or even for the entire library data structure,
beginning on the level from where the validation starts. For further details on
the single reports, see section 7.5.1.
If you perform more than one validation, the result files will be consecutively
numbered (*.v00, *.v01...). You can compare different validation results with
each other by selecting the respective file from the drop-down list. You can also
print the report by clicking on the Print button. This starts the Windows Notepad
program which you can use to reformat the text, if desired. Use the Notepad
print function to create a printout.
Bruker Optik GmbH
OPUS/IDENT
15
Setting Up an Identity Test Method
Note: It is recommended to select a small font in Windows Notepad to avoid
extraordinary long reports. A proportional font may lead to a confusing display of
the results. Therefore, it is advisable to use a monospace font, e.g. Courier New,
10.
The Threshold values listed on the Threshold tab are regarded as confidence
region during validation. The results are classified in three categories: uniquely
identified, not identified and can be confused with. In case of results belonging
to the first category the spectral distance between the original and average
spectrum is within the threshold value. The spectral distance is higher than the
threshold value in case of results belonging to the second category. The Can be
confused category indicates that the spectral distance of an original spectrum is
smaller than the confidence level, compared to at least one different average
spectrum. For further details, see section 6.7.
1.6
Storing Method Files
Click on the Store Method tab to store the method files you have created. The
following dialog opens:
Figure 13: Setup Identity Test Method - Store Method tab
16
OPUS/IDENT
Bruker Optik GmbH
Storing Method Files
The parameter you can define in this tab is the Number of Hits in Ident Report,
i.e. you enter the number of hits that have to be stored together with the IDENT
report.
By default, the Store Average Spectra check box is activated. This means that
you can store all average spectra of a library (IP1 level) in one separate
directory which is a sub-directory of the directory in which the IDENT method
has been stored.
The average spectra will be stored without data pre-processing, and have an
AVERAGE (
) data block appended. If you repeatedly store a particular
IDENT method, the average spectra will NOT be overwritten. Instead, the file
extension will be incremented.
Click on the Store Method button. The standard Save File dialog box opens to
be used to save the method. The method file uses the extension *.FAA, and all
sub-libraries will be stored simultaneously.
For special details on method protection refer to the OPUS QUANT manual,
chapter 9.
Bruker Optik GmbH
OPUS/IDENT
17
Setting Up an Identity Test Method
18
OPUS/IDENT
Bruker Optik GmbH
2
Performing an IDENT
Analysis
Compared to the setup of an IDENT method the analysis of unknown samples is
easy. Before you start the analysis load the spectra of your unknown samples
into the OPUS browser window.
The analysis compares the test spectrum with all reference spectra. The result of
a comparison between spectrum A and B results in the spectral distance D,
which is also called Hit Quality. The better two spectra match, the smaller the
spectral distance. The Hit Quality for identical spectra is 0 (i.e. a reference
spectrum is compared with itself).
To start an IDENT analysis, select the Identity Test command from the Evaluate
menu. The Identity Test dialog box opens.
Figure 14: Identity Test - Select File(s)
Drag and drop the absorption block of the test spectra from the OPUS browser
window into the File(s) for Identity Test field. If you release the left mouse
button, the spectra are added to this entry field.
Bruker Optik GmbH
OPUS/IDENT
19
Performing an IDENT Analysis
If an identity test method has already been loaded (e.g. if you have created a
method prior to starting the analysis), the path and name of this method name
will be indicated in the Loaded Identity Test Method field. To load or change an
IDENT method, click on the Load Ident Method button and select the desired
method from the dialog box that opens.
The analysis uses the No Reference Defined function for an IDENT method if
you have not defined an expected reference. Click on the Change button to
modify this default setting. The Change button will be disabled if the operator
has no right to change parameters of this kind. For details on this dialog, see
section 7.7.1.
Click on the Identity Test button to start the test. The result of the analysis is
appended to the respective file in the form of an IDENT report block. If you
click on this report block, a report window opens automatically and displays the
results.
20
OPUS/IDENT
Bruker Optik GmbH
Identity Test Reports
3
IDENT Report
Both the results of an identity test and the averaging of original spectra to
generate reference files for an IDENT library are stored in report blocks. You
can open these reports, like any other OPUS report, by double-clicking on the
report block in the OPUS browser window.
3.1
Identity Test Reports
The results of the spectrum comparison between test spectrum and reference
spectra are written into an IDENT report. This report is stored in the test
spectrum file. The content of the report depends on the parameters and
algorithms selected to run the identity test. For details, see section 6.4. If you
click on the REPORT data block, a text window will open and show the results.
3.1.1
Standard Method
The IDENT report contains detailed information on the method and a list of
spectral distances between the test spectrum and the reference spectra. This list
includes distances in ascending order, i.e. Hit No. 1 is the reference spectrum
which is most similar to the test spectrum. The number of listed distances can be
defined when creating the IDENT method (see chapter 7.6).
Figure 15 shows an IDENT report using the Standard algorithm in combination
with an identity test on the first reference spectrum (L-Leucin) of the library. An
original (individual) spectrum of the reference library has been used as test
spectrum. The threshold of the selected reference spectrum (Threshold for
expected reference, see figure 15) is 0.023866. The spectral distance between
the test spectrum and this reference spectrum is 0.010556 (Hit quality with
expected reference), i.e. it is smaller than the threshold. Since no further hit can
be found below this threshold (Hit No. 2 with 0.047006 is higher), the result is:
IDENTICAL TO to the expected spectrum.
Bruker Optik GmbH
OPUS/IDENT
21
IDENT Report
Figure 15: IDENT Report – Query spectrum identical
Figure 16 shows a test report for the same query spectrum. But this time
DL-Methionin has been selected as expected reference. Now, the result is NOT
IDENTICAL to the expected spectrum. The spectral distance to this reference
spectrum is 0.172990 and Hit No. 7. This value exceeds the threshold of
0.072976, and therefore the test spectrum is classified as not being identical.
22
OPUS/IDENT
Bruker Optik GmbH
Identity Test Reports
Figure 16: Ident Report – Query spectrum not identical
3.1.2
Factorization Method
If you use the Factorization algorithm (on the Parameters tab), Eigen values
and Eigen vectors (see section 6.1.2) may also be interesting for you. The factor
values can be retrieved from the report file of the IDENT method file. Load the
method file (extension *.FAA) into the OPUS browser window and click on the
REPORT data block to open the report. Open the Identity Search Method
subdirectory as shown in figure 17. If you click on Eigen Vectors or Eigen
Values, these values will be displayed in the report window. The T values (see
section 6.1.2) are listed in the Eigen vectors sub-directory.
Bruker Optik GmbH
OPUS/IDENT
23
IDENT Report
Figure 17: Report file of a method file using factorization
24
OPUS/IDENT
Bruker Optik GmbH
Theory
4
Cluster Analysis
The cluster analysis tests FT-IR spectra for their similarity. In contrast to the
identity test, no input information is required. The cluster analysis divide
similar spectra into groups. These groups are called classes or clusters. The
clustering can be displayed in a dendrogram. Figure 18 shows a simplified
dendrogram including 5 spectra.
A
C
B
D
E
Figure 18: Cluster Analysis – Dendrogram
4.1
Theory
The spectral distance indicates the degree of spectral similarity. Two spectra
with a spectral distance of 0 are entirely identical (within the frequency ranges
tested). The higher the difference between two spectra, the higher the spectral
distance.
Bruker Optik GmbH
OPUS/IDENT
25
Cluster Analysis
The hierarchical cluster algorithms perform the following tasks:
• First, the spectral distances between all spectra are calculated.
• The two spectra with the highest similarity (i.e. spectra with the
smallest spectral distance) are merged into a cluster.
• The distances between this cluster and all other spectra are
calculated. Several methods (Single Linkage, Complete Linkage...)
are available to calculate the distances.
• The two spectra (spectrum/spectrum or spectrum/cluster) with the
smallest distance are merged again into a new cluster.
• The distances between this new cluster and all other spectra (spectra,
cluster) are calculated.
• The two spectra (spectrum/spectrum or spectrum/cluster or cluster/
cluster) are merged into a new cluster.
...
This procedure will be repeated until only one big cluster will be left.
Figure 19 shows this procedure in more detail. The spectral distances between
any two spectra of a set of n spectra can be represented in a n × n matrix. This
matrix is symmetrical and the main diagonal elements are 0. Subsequently, it is
sufficient to test only a triangle of this matrix.
Five spectra A, B, C, D and E are used in the example. The triangular matrix
which contains the spectral distances between these 5 spectra is shown in the
upper part of figure 19. The A and C spectra are mostly similar to each other.
The spectral distance is 11.0. Both spectra are merged into the AC cluster.
26
OPUS/IDENT
Bruker Optik GmbH
Theory
A
B
C
D
E
A
0
44.0
11.0
101.6
118.3
AC
B
D
E
AC
0
49.2
99.5
117.1
AC
B
DE
AC
0
49.2
108.3
ABC
DE
ABC
0
94.2
B
C
D
E
0
54.4
68.1
92.1
0
97.4
115.9
0
21.2
0
B
D
E
0
68.1
92.1
0
21.2
0
B
DE
0
80.1
0
DE
0
Figure 19: Spectral distances calculated using a hierarchical cluster method
Now, the distances between AC and the other spectra have to be calculated. This
time the Average Linkage algorithm has been used. The distance between A and
D is 101.6 and the distance between C and D is 97.4. The distance between the
AC cluster and D spectrum is the mean value of the two original distances:
(101.6 + 97.4) / 2 = 99.5. The new distance values can be seen in the second
matrix of figure 19.
As the smallest spectral distance is 21.2 in this matrix, the D and E spectra are
merged into the DE cluster. Then, the distances between this new cluster and all
other spectra will be calculated again.
Example: The spectral distance between AC and D is 99.5 and between AC and
E is 117.1. Based on these values, the distance between AC and DE will be
108.3. The third matrix in figure 19 includes these new distance values.
Bruker Optik GmbH
OPUS/IDENT
27
Cluster Analysis
In the next step, the B spectrum is merged with the AC cluster into a new ABC
cluster. The distance between the ABC and DE cluster is 94.2. Finally, the ABC
and DE clusters are merged into the ABCDE cluster.
The y axis of the dendrogram shows the spectral distances between different
clusters. The horizontal lines indicate the fusion levels, which are the spectral
distances of the different clusters and spectra prior to new clustering.
Table 1: Clustering Process
Number of Clusters
Clusters
5
A – B – C– D – E
4
AC – B – D – E
3
AC– B – DE
2
ABC– DE
1
ABCDE
You have to generate a cluster analysis method before a dendrogram can be
graphically displayed. While you generate the method the spectral distances
between the different spectra are calculated.
The clustering is repeated until all spectra are merged in one single cluster.
Sometimes the intermediate states are of great interest for the user. If you use
the Make Diagnosis function, you can get a cross section of the dendrogram.
Simply enter the number of classes, and a list including the components of each
single class will be generated. Additionally, the spectral distance for each
cluster is indicated which has recently been merged.
As already mentioned, the spectral distances between different spectra can be
represented by a symmetrical matrix. If you use the Make Histogram function,
the whole matrix or part of it can be statistically tested. The results will be
shown in the form of a histogram. However, this part of the program does not
include clustering in general, but calculates the distances between the different
spectra only.
4.1.1
Methods to Calculate Spectral Distances
Four different methods can be used to calculate spectral distances:
•
•
•
•
Standard algorithm
Factorization
Scaling to 1st Range
Normalize to Reprolevel
These methods are an integral part of the IDENT software. For details, see
chapter 6 and 7. The Standard algorithm uses the Euclidian distance to
determine spectral distances, while Scaling to 1st Range and Normalize to
Reprolevel use the correlation coefficient.
28
OPUS/IDENT
Bruker Optik GmbH
Theory
Using the Standard or Factorization method the spectral distances calculated by
the cluster analysis differ from those calculated by the identity test. Overlapping
frequency ranges will not be merged when using cluster analysis. An artificial
spectrum is derived from the selected spectral ranges of the measured spectrum.
The artificial spectrum is used for the calculation of the spectral distances (and
the data preprocessing) and includes numerous data points of the overlapping
frequency regions.
The Scaling to 1st Range and Normalize to Reprolevel algorithms separately
calculate the spectral distances for each frequency range. Then, an average
value is calculated, and each frequency range can be weighted differently. If
you use Normalize to Reprolevel, you can additionally specify a reproduction
level for each frequency range. This level can be determined by the Make
Histogram function.
The calculated spectrum-to-spectrum distances of the cluster analysis are equal
to those calculated by the identity test if you use the Normalize to Reprolevel
method. This, however, does not apply to Scaling to 1st Range. The identity test
uses the spectral distances between the test spectrum and n reference spectra to
determine extrema. This means that n distance values have to be taken into
account per each frequency range. The cluster analysis, however, uses the
distances between all reference spectra to determine extrema. These are
( n ⋅ ( n – 1 ) ) ⁄ 2 distances for n reference spectra as the spectral distance of a
reference spectrum to itself is not considered.
To be able to compare the results achieved by the identity test and cluster
analysis using the Normalize to Reprolevel and Scaling to 1st Range methods,
the parameters required for the identity test and cluster analysis must be
identical (same reference spectra, same frequency ranges etc.). Then, start the
identity test. Use the first or last reference spectrum as test spectrum. Compare
the Hit Qualities of this IDENT test report with the spectrum-to-spectrum
distances of the cluster analysis.
4.1.2
Cluster Algorithms
There are 7 methods available to calculate spectral distances between a newlycreated cluster and all the other spectra or clusters. The algorithms most
frequently used are Average Linkage and Ward’s Algorithm.
Single Linkage
The p and q clusters are merged to the new r cluster. D(p,i) is the spectral
distance between the p and i clusters, while D(q,i) is the spectral distance
between the q and i clusters. The D(r,i) distance between the new r cluster and
the i cluster is the smaller one of the two original distances:
D ( r, i ) = min [ D ( p, i ), D ( q, i ) ]
(4-1)
This method can be used to create large clusters.
Bruker Optik GmbH
OPUS/IDENT
29
Cluster Analysis
Complete Linkage
The new distance is the larger one of the two original distances.
D ( r, i ) = max [ D ( p, i ), D ( q, i ) ]
(4-2)
This method prefers to create small groups.
Average Linkage
The arithmetic mean value is calculated as follows:
( p, i ) + D ( q, i )
D ( r, i ) = D
---------------------------------------2
(4-3)
Weighted Average Linkage
n(p) is the number of spectra which are merged in the p cluster and n(q) is the
number of spectra which are merged in the q cluster. The spectral distance
between the new r cluster and the i cluster is calculated as follows:
( p ) ⋅ D ( p, i ) + n ( q ) ⋅ D ( q, i -)
D ( r, i ) = n----------------------------------------------------------------------n( p) + n(q )
(4-4)
This algorithm is a generalization of Average Linkage.
Median Algorithm
D(p,q) is the spectral distance between p and q.
( p, i ) + D ( q, i ) – D
( p, q -)
D ( r, i ) = D
-------------------------------------------------------2
4
(4-5)
Centroid Algorithm
n is the total number of reference spectra. D(r,i) is calculated according to the
following equation:
n ( p ) ⋅ D ( p, i ) + n ( q ) ⋅ D ( q, i )- + n--------------------------------------------------( p ) + n ( q ) ⋅ D ( q, p )
D = ----------------------------------------------------------------------2
n
n
(4-6)
Ward’s Algorithm
The previous algorithms merge the two groups which are most similar. Ward's
Algorithm, however, tries to find as homogeneous groups as possible. This
means that only two groups are merged which show the smallest growth in
heterogeneity factor H. Instead of determining the spectral distance, the
Ward’s Algorithm determines the growth of heterogeneity H. This method can
30
OPUS/IDENT
Bruker Optik GmbH
Performing a Cluster Analysis
especially be used for the cluster analysis of bacteria spectra, i.e. these clusters
correlate extremely well with microbiological affinity.
n(i) is the number of spectra merged in the i cluster. H(r,i) is calculated
according to the following equation:
n ( p ) + n ( i ) ] ⋅ D ( p, i ) + [ n ( i ) + n ( q ) ] ⋅ D ( q, i ) – n ( i ) ⋅ D ( q, i )- (4-7)
H ( r, i ) = D ( r, i ) = [------------------------------------------------------------------------------------------------------------------------------------------------------n + n(i)
If you want to test different cluster algorithms, you do not need to calculate the
spectrum-to-spectrum distance matrix again, because the clustering does not
have any effect on this kind of matrix.
4.2
Performing a Cluster Analysis
To perform a cluster analysis the following steps are required:
• Measuring at least 1 spectrum per substance
• Incorporating the spectra into the list
• Defining a suitable spectral range for identification
• Selecting a data preprocessing method
• Calculating spectral distances
• Defining a cluster algorithm
• Generating a dendrogram, histogram or diagnosis
Several tabs of the Cluster Analysis dialog box are identical to the Setup Identity
Test Method dialog and will only be briefly explained. For details, see the
respective sections in chapter 1.
Click on the Cluster Analysis command from the Evaluate menu.
On the Load Method tab you can load an already existing method file generated
for cluster analysis. This tab is identical to the Load Method tab of the Setup
Identity Test Method command. As you will generate a new Cluster Analysis
method, click on the Reference Spectra tab. The following dialog opens:
Bruker Optik GmbH
OPUS/IDENT
31
Cluster Analysis
Figure 20: Cluster Analysis – Reference Spectra tab
To create a list, first define the spectra to be used on the Reference Spectra tab.
Add these spectra as described in chapter 1.
Now, click on the Parameters tab. Define the spectral regions to be considered
for cluster analysis and select a data preprocessing method as well as the cluster
analysis algorithm.
In case of data preprocessing you can select between Vector Normalization,
First and 2nd Derivative, as well as combinations of both. Vector
Normalization is set by default.
32
OPUS/IDENT
Bruker Optik GmbH
Performing a Cluster Analysis
Figure 21: Cluster Analysis – Parameters tab
Define the frequency regions you want to use, see chapter 1. Select the
Standard method to calculate the spectral distance.
Click on the Calculate Distances button. After the calculation of the distances
you first have to save the method. Click on the Store Method tab and on the
Store Method button to open the standard Save File dialog box. Enter a new file
name for the method file and click on the Save button.
Bruker Optik GmbH
OPUS/IDENT
33
Cluster Analysis
Figure 22: Cluster Analysis – Store Method tab
Now, click on the Report tab to have the results displayed. You can select
between different display views.
34
OPUS/IDENT
Bruker Optik GmbH
Performing a Cluster Analysis
Figure 23: Cluster Analysis – Report tab with dendrogram
If you click on the Window button, you can modify the report display. A new
dialog will open which allows to change the algorithm and the kind of labeling.
Bruker Optik GmbH
OPUS/IDENT
35
Cluster Analysis
Figure 24: Cluster Analysis - Options to modify report display
To return to the report click on the Back to Report button.
The Options button on the Report tab also enables to modify the report display,
and to define the algorithm and matrix parameter in more detail. You can select
Sample Name from the Dendrogram drop-down list in the Cluster Analysis
dialog to change the dendrogram labeling accordingly. You can also add file
names, sample names or file numbers to the dendrogram or have the
dendrogram displayed without any labeling.
36
OPUS/IDENT
Bruker Optik GmbH
3D Files/Filelist
Figure 25: Cluster Analysis – Options
For further details on the Cluster Analysis dialog, see section 7.11.
4.3
3D Files/Filelist
Instead of single spectra it is also possible to use 3D files or file lists in
connection with the cluster analysis. Note that not more than one 3d file or file
list can be loaded to perform a cluster analysis. For further details on this
subject refer to chapter 7.12.
Bruker Optik GmbH
OPUS/IDENT
37
Cluster Analysis
38
OPUS/IDENT
Bruker Optik GmbH
Setting up Conformity Test
5
Conformity Test
The conformity test is a an easy method to test the deviations of measured NIR
spectra within certain limits. To set these limits you need samples, which
belong to at least one batch or one production cycle, of the final product as
reference spectra. These reference spectra vary within the accepted range of
specifications. The NIR spectra of these samples reflect the different sample
variations and form a confidence band in the spectral range. To pass the
conformity test, the spectrum of a new sample has to be within this confidence
band on each wavelength.
The conformity test is mainly used for the quality control of defined products
for which a quantitative calibration would be too time-consuming or even
impossible.
First, you have to calculate the average and the standard deviation σ of the
absorbance values for each wave length i. The mean value plus/minus the
standard deviation determine the confidence band within the spectral range, and
define which amount of variations on each spectral wavelength is acceptable for
the particular product.
Second, you have to check whether the spectrum of a sample to be tested is
within the defined confidence band in the spectral range. The difference
between this sample and the average of the reference samples is calculated on
each wave length i. This absolute deviation is now weighted by the
corresponding standard deviation σ on the respective wavelength, which results
in a relative deviation referred to as Conformity Index (CI).
CI = (Areference,i - Asample,i)/ σreference,i
The maximum of all CI values is derived as test result.
5.1
Setting up Conformity Test
Select the Setup Conformity Test command in the Evaluate menu. On the Load
Method tab you can load an already existing method file generated for a
conformity test.
Bruker Optik GmbH
OPUS/IDENT
39
Conformity Test
Figure 26: Setup Conformity Test - Load Method tab
The General information group field includes information on the kind of data
block, data points selected as well as the data preprocessing type. It is
distinguished between reference and test spectra. Reference spectra have been
created by a specific method, whereas test spectra can be tested for their
conformity with this specific method for validation purposes.
To create a new method click on the Reference Spectra tab and load the
respective reference spectra by clicking on the Add Reference Spectra button. A
dialog pops up and displays a browser window which you have to use to search
for and select the spectra. Click on the Open button to load the spectra.
40
OPUS/IDENT
Bruker Optik GmbH
Setting up Conformity Test
Figure 27: Setup Conformity Test - Reference Spectra tab
The Reference Spectra and Test Spectra tab are based on the same principle.
Figure 27 exemplifies a method which consists of a large number of reference
spectra. The Data Set column specifies the type of spectra, in this case R
indicates reference, T test. Further spectra features are the path, file and sample
name, which are the same for both tabs.
If you want to change the path for the conformity test spectra, click on the
Change Path tab. A dialog opens which you use to define the new path.
It is also possible to modify the data set. Select the respective reference or test
spectra and click on the Set Data Set button.
Bruker Optik GmbH
OPUS/IDENT
41
Conformity Test
Figure 28: Setup Conformity Test - Set Data Set option
You can temporarily exclude some spectra or change them from reference to
test spectra (or vice versa), or restrict the data set to a few spectra (Excluded
option) only and create a new method. As the quality of a method depends on
the reference and test spectra, make sure that the data sets are carefully
assembled.
Select one of the options from the drop-down list and click on the Set Data Set
button. To continue click on the Exit button.
The spectra excluded will be marked in gray, see figure 29.
Figure 29: Setup Conformity Test - Spectra list with changed data set
On the Parameters tab you define the spectral regions to be used for the
conformity test and select a data preprocessing method.
42
OPUS/IDENT
Bruker Optik GmbH
Setting up Conformity Test
Figure 30: Setup Conformity Test - Parameter tab
Frequently, the Vector normalization is selected as preprocessing method.
Sometimes better results can be obtained by using the First or 2nd derivative
method. In both cases you have to additionally define the amount of Smoothing
points, you can select between 5 (9) and 25. The optimal number of smoothing
points, however, has to be evaluated empirically.
The Conformity Index Limit parameter records spectra between an upper and
lower limit (see graph below). The best possible scaling factor is between 3 and
4 which is, of course, not mandatory. The evaluation basis for the self-adapting
conformity test is the reference spectra scaling, indicated by means of the
standard deviation.
Upper CI limit
Lower CI limit
Bruker Optik GmbH
OPUS/IDENT
43
Conformity Test
Activate the Use signed Conformity Index Values check box if you want to use
non-absolute index values.
Click on the Display Preprocessed Spectra button to have the respective spectra
and confidence band displayed. The confidence band is displayed in red and is
calculated as follows:
Average value ± CI limit • standard deviation
Figure 31: Preprocessed spectra plot
Use the selection box on the lower part of the dialog (figure 31) to have the
spectra selectively displayed. Deactivate the Show check box of the spectra
which you do not want to have displayed. If you click on the Interactive Region
Selection button, the standard Select Frequency Range(s) dialog opens which
you can use to interactively set the frequency range. To continue with the
conformity test click on the Go to Setup Conformity Test button.
If you have defined all the parameters required, click on the Validate tab. This
tab only includes the Validate button. Click on this button, and you will
automatically be transferred to the Graph tab.
44
OPUS/IDENT
Bruker Optik GmbH
Setting up Conformity Test
A
B
C
Figure 32: Setup Conformity Test - Graph tab
The green data points (A) represent the reference spectra, and the blue ones (B)
represent the test spectra. The CI Limit is indicated by the red line (C) which
can be moved by using the CI limit slider on the left side. To display the
reference and test spectra separately, select either Reference or Test from the
upper drop-down list. In most cases it is recommended to select the Reference +
Test option, to be able to directly compare the scattering of reference vs test
spectra.
If you position the cursor on one specific data point, a text frame pops up
indicating the exact spectrum identity (see figure 33).
Figure 33: Spectra plot with spectrum description
You can also select single data points only. Move the cursor to the respective
data point section, press the left mouse button and draw the mouse over this
section. If you leave the mouse button, only the data points selected will be
displayed in the plot. To undo this, just right click into the plot.
Bruker Optik GmbH
OPUS/IDENT
45
Conformity Test
You can select between the following algorithms:
• Max Conformity Index
The maximum value will be calculated based on the frequency
ranges selected.
• Sum 1: Sum over CI > Limit (/N total)
All y-values above the CI limit are added up and divided by the total
number of data points within the frequency ranges selected.
• Sum 2: Sum over CI > Limit (/N over Limit)
All y-values above the CI limit are added up and divided by the
number data points which are above the CI limit.
Depending on your specific quality control problem, you select either one of
these algorithms. General recommendations cannot be made, you have to
empirically find out which procedure would be the best for your specific
requirements.
If you click on the Display CI Spectra button, the CI spectra and the CI limit
will be displayed.
Figure 34: Setup Conformity Test - CI spectra
The reference spectra are displayed in blue, and the test spectra are displayed in
green. To localize a specific spectrum and analyze the standard deviation,
deactivate and re-activate the Show check box. Click on the Go to Setup
Conformity Test button to return.
46
OPUS/IDENT
Bruker Optik GmbH
Setting up Conformity Test
To display the spectra report click on the Report tab.
Figure 35: Setup Conformity Test - Report tab
Depending on the algorithm selected from the drop-down list, the
corresponding outliers are marked in grey color.
The Data Points >CI Limit column refers to the data points above the CI limit,
which is important to analyze the integral. The values displayed in the Sum 1
column represent the integrals divided by all data points, and the values
displayed in the Sum 2 column represent the integrals divided by the data points
which are above the CI limit. If you want to print the report, click on the Print
button.
Before you store the method, you have to define the analysis method. Either
activate the Use CI Limit, Use Sum 1 Limit or Use Sum 2 Limit option button on
the Store Method tab.
Bruker Optik GmbH
OPUS/IDENT
47
Conformity Test
Figure 36: Setup Conformity Test - Store Method tab
5.2
Performing Conformity Test
Start the conformity test by clicking on the Conformity Test command. The
following dialog opens:
48
OPUS/IDENT
Bruker Optik GmbH
Performing Conformity Test
Figure 37: Conformity Test - Select Files tab
Drag & drop the file(s) to be evaluated from the OPUS browser window into the
File(s) for Conformity selection field.
Click on the Load Conformity Method button and load the particular method
which path is displayed above this button.
The Conformity Test results will be stored in a CONF data block (
displayed in a specific report view.
) and
Figure 38: Conformity Test Report
Bruker Optik GmbH
OPUS/IDENT
49
Conformity Test
50
OPUS/IDENT
Bruker Optik GmbH
Algorithms
6
IDENT Theory
The aim of an IDENT analysis is to determine the differences between a test
spectrum and the reference spectra of a library. You have to define a method to
test the similarity of spectra, and a threshold. This threshold determines whether
a spectrum is only similar or even identical to the reference spectrum.
6.1
Algorithms
Basically, there are two algorithms to perform the IDENT analysis: the
Standard and Factorization method. During the analysis both methods compare
the test spectrum with all reference spectra. The result of a comparison between
two spectra is the Hit Quality, also referred to as spectral distance D. The better
two spectra match, the smaller the spectral distance. The Hit Quality for
identical spectra is 0 (i.e. if a reference spectrum is compared with itself).
6.1.1
Standard Method
Figure 39 shows two spectra a and b, one test and one reference spectrum. The
spectral distance D is proportional to the area between these two curves. The
following formula for the so-called Euclidean distance is used in the Standard
method:
D =
∑
(a(k) – b(k))
2
(6-2)
k
where a(k) and b(k) are the ordinate values of the a and b spectra. The sum
incorporates all selected k data points.
In the current IDENT report (see figure 40) the smallest spectral distance which
has been determined by spectrum comparison is 0.96. In this case one of the
sample spectra previously used to generate an average spectrum for the
reference library serves as a test sample, which explains the extremely small
spectral distance. As this example shows, the first two hits are well separated
from each other. The spectral distance for Hit No. 2 is 6.19, which is about 6
times the smallest spectral distance.
Bruker Optik GmbH
OPUS/IDENT
51
IDENT Theory
a
b
Figure 39: Two spectra and their spectral distance
52
OPUS/IDENT
Bruker Optik GmbH
Algorithms
Figure 40: Ident Report
6.1.2
Factorization
The Factorization method represents spectra as linear combinations of so-called
factor spectra (loadings):
a = T 1a ⋅ f 1 + T 2a ⋅ f 2 + T 3a ⋅ f 3 + …
(6-3)
The a vector shows the a spectrum and the factor spectra are denoted f1, f2, f3
etc. T indicates the coefficients (scores) required to reconstruct the original
a spectrum.
To calculate the spectral distance D between the two spectra a and b, the T
coefficients are used in the Factorization method:
D =
Bruker Optik GmbH
∑i ( Tia – Tib )
OPUS/IDENT
2
(6-4)
53
IDENT Theory
The summation is performed for a certain number of coefficients. These T
coefficients are also called scores. The differences between the original and
reconstructed spectrum are known as spectral residuals (figure 41).
Figure 41: Reference spectrum, reconstructed reference spectrum and difference spectrum
The Standard method directly uses the spectral intensities to calculate the
spectral distance, and the summation is performed for all data points within the
specified frequency regions (which could be more than 1000 points). How
many factor spectra or score coefficients have to be included in an IDENT
library is a very important aspect and will be explained in the following.
When factorizing an IDENT library, s average spectra are transformed into
s factor spectra. These factor spectra are orthogonal to each other. The effect a
certain factor has on the reproduction of reference spectra is indicated by the
respective Eigen value. The factor spectra are sorted according to these
Eigen values. The first factor spectrum is the most important one and thus has
the highest Eigen value.
The more the Eigen value decreases, the lower the spectral intensities (ordinate
values) of the factor spectra, and the more intensive the noise. Factor spectra
which mainly consist of noise must not be used for an IDENT method.
Factor spectra are stored in the IDENT method directory, using the OPUS file
format. They have the same file name as the method. However, the spectra have
an additional numeric file extension, starting with 0 for the first factor spectrum.
These spectra can be loaded into OPUS like any other spectrum.
54
OPUS/IDENT
Bruker Optik GmbH
Algorithms
Figure 42: Reference spectrum and first factor spectrum
Figure 43: Second, third and forth factor spectrum
The spectra (see figure 42 and 43) show the signal-to-noise ratio of a certain
factor spectrum. The factor spectrum displayed in figure 44 mainly consists of
noise. It is not recommended to use this spectrum to calculate spectral distances.
Bruker Optik GmbH
OPUS/IDENT
55
IDENT Theory
You can easily check the factor spectra orthogonality by multiplying two factor
spectra using the OPUS Spectrum Calculator. This is followed by an integration
across the whole frequency range of the result spectrum. The integration result
will be 0 (or approximately 0 due to the finite computing accuracy).
Figure 44: Factor spectrum with excessive noise
6.2
Factorization Theory
Assuming that s reference spectra consist of d data points each. The reference
spectra are represented by d1, d2, d3 ... column vectors which form D data
matrix ( d × s dimension):
D = [ d1 d2 d3 … ds ]
(6-5)
When exchanging the rows for the columns in this matrix, you obtain
DT transposed matrix of s × d dimension. Multiply DT transposed matrix by
D original matrix to obtain Z covariance matrix:
T
Z = D ⋅D
(6-6)
A diagonalization and orthogonal transformation of Z produce Eigen vectors
and Eigen values of Z.
T
Λ = L ⋅Z⋅L
56
OPUS/IDENT
(6-7)
Bruker Optik GmbH
Factorization Theory
The column vectors of L matrix ( s × s dimension) are l1, l2, l3... Eigen vectors
of Z matrix. These Eigen vectors are orthonormal, i.e. the following equations
are valid for the scalar product of two Eigen vectors:
li ⋅ lj = 0
i≠j
(6-8)
li ⋅ li = 1
(6-9)
Λ matrix ( s × s dimension) contains λ1, λ2, λ3 ... λs Eigen values of Z matrix as
the main diagonal, all other matrix elements are 0. This means:
Z ⋅ li = λi ⋅ li
(6-10)
D data matrix is factorized by L Eigen vector matrix, using multiplication:
F = D⋅L
(6-11)
F matrix has the same dimensions as D data matrix ( d × s ) and includes the
vectors of f1, f2, f3, ... factor spectra as columns. Multiplying FT transposed
matrix by F yields:
T
T
T
T
T
F ⋅ F = (D ⋅ L) × (D ⋅ L) = L ⋅ D ⋅ D ⋅ L = L ⋅ Z ⋅ L = Λ
(6-12)
The elements of FTF quadratic matrix ( s × s dimension) are the scalar products
which can be created in pairs together with factor spectra. The 6-12 equation
causes the factor spectra to be orthogonal to each other:
fi ⋅ fj = 0
i≠j
(6-13)
fi ⋅ fi = λi
(6-14)
The vector norm of a factor spectrum is equal to the square root of the
corresponding Eigen value. Using L orthogonality, D data matrix can be as
follows:
T
D = D⋅1 = D⋅L⋅L = F⋅L
T
(6-15)
The reference spectra are represented as linear combinations of the factor
spectra, and the coefficients are contained in the columns of LT matrix. Based
on the 6-15 equation the following applies to the first reference spectrum:
T
T
T
T
d 1 = L 1, 1 ⋅ f 1 + L 2, 1 ⋅ f 2 + L 3, 1 ⋅ f 3 + … + L s, 1 ⋅ f s
(6-16)
The score coefficients are the coordinates of the reference spectra in the factor
spectra system.
Any u spectrum can be represented as linear combination of the factor spectra:
u = F⋅k⋅e
Bruker Optik GmbH
OPUS/IDENT
(6-17)
57
IDENT Theory
The unknown k column vector corresponds to the column elements of
LT matrix. E error spectrum is the difference between the u spectrum and
reconstructed spectrum. The Least Squares solution for k which minimizes the
error is as follows:
T
k = (F ⋅ F)
–1
T
⋅F ⋅u = Λ
–1
T
⋅F ⋅u
(6-18)
If only the first r factor spectra are taken into account, D spectral distance
between one u spectrum and one da reference spectrum is:
T
2
T
2
T 2
( k 1 – L 1a ) + ( k 2 – L 2a ) + … + ( k r – L ra )
D =
(6-19)
Instead of using the column vectors of LT matrix you can use L row vectors. The
equation for D spectral distance in r dimensional factor space is then:
2
2
( k 1 – L a1 ) + ( k 2 – L a2 ) + … + ( k r – L ar )
D =
2
(6-20)
When using the Standard method you can specify a range of values for
D distances. Select Vector Normalization preprocessing. The range of values
reaches from 0 (identical spectra) to 2 (maximum spectral difference). This does
not apply to D when using the Factorization method. If all factor spectra are
used, the spectral distances between reference spectra are constant, i.e.
D = 2 . To calculate spectral distances, the elements of L Eigen vector matrix
are used. This matrix consists of orthogonal unit vectors in s dimensional space.
The distance between two orthogonal unit vectors always has to be 2 .
However, when using the Factorization method not all factor spectra are used,
because the higher factor spectra mainly increase the noise level of the
reconstructed spectrum.
The Factorization method also allows to calculate spectral distances by using
residuals. For details see chapter 7.3. The spectral residual is calculated from
the difference between the original and reconstructed spectrum. To calculate
SpecResu spectral residual (u being an arbitrary spectrum) the equation is as
follows:
SpecRes u =
∑
( u ( k ) – ( k1 ⋅ f1 ( k ) + k2 ⋅ f2 ( k ) + … + kr ⋅ fr ( k ) ) )
2
(6-21)
k
The summation is performed for all selected k data points.
D spectral distance between u spectrum and da reference spectrum is calculated
as follows:
D =
2
2
2
( k 1 – L a1 ) + ( k 2 – L a2 ) + … + ( k r – L ar ) + ( SpecRes u – SpecRes a )
2
(6-22)
Figure 45 shows a scheme representing the Factorization method.
58
OPUS/IDENT
Bruker Optik GmbH
Factorization Theory
Reference Spectra D
Factorization
Factor Spectra F
Eigen values Λ
Coefficients
LT
Spectral Distance
Coefficients k
Test Spectrum u
Figure 45: Factorization – Spectral distance calculation
6.2.1
Scaling to 1st Range and Normalize to
Reprolevel
Scaling to 1st Range and Normalize to Reprolevel are algorithms that can be
used to identify microorganisms. Contrary to the Standard and Factorization
method overlapping spectral ranges are not merged. For example, if you set
1500-1200cm-1 as first frequency range and 1500-1400cm-1 as second
frequency range, they will not be combined into one frequency range, and the
data points in the 1500-1400cm-1 range will be considered twice to calculate
spectral distances. The single spectral ranges can be weighted by different
factors. These factors are defined in the Weight column (see chapter 7.3).
The Vector Normalization preprocessing is not available in combination with
the Scaling to 1st Range and Normalization to Reprolevel algorithms, as a vector normalization will be automatically performed in this case. Contrary to the
Bruker Optik GmbH
OPUS/IDENT
59
IDENT Theory
Standard and Factorization algorithm the vector normalization is calculated
separately for each spectral range. The resulting values are used to determine a
mean distance. The vector normalization considers all data points selected,
when using the Standard and Factorization algorithm.
Calculating spectral distances
1) To calculate spectral distances spectra have to be vector normalized
first, for each frequency range.
2) Then, the r Pearson correlation coefficient is calculated. This kind of
coefficient defines the correlation between a and b spectra:
r =
∑ an ( k ) ⋅ bn ( k )
(6-2)
This correlation is calculated separately for each frequency range. The
summation covers all k data points of a frequency range: an and bn are
the normalized spectral intensities. The normalization yields:
∑ ( an ( k ) – am ) ⋅ ( bn ( k ) – bm )
r = --------------------------------------------------------------------------------------------2
∑ ( an ( k ) – am ) ⋅ ∑ ( bn ( k ) – bm )2
(6-3)
am and bm are the mean spectral intensities within the spectral range,
while a(k) and b(k) are the original spectral intensities.
The value range of r correlation coefficient reaches from -1 (inverse
spectra) to +1 (identical spectra).
3) The correlation coefficient is transformed into D spectral distance by the
following equation:
D = ( 1 – r ) ⋅ 1000
(6-4)
D spectral distance can be between 0 (identical spectra) and 2000
(inverse spectra).
4a) The Scaling to 1st Range determines the minimum and maximum value
of spectral distances for the first spectral range. Then, the distances of all
the other spectral ranges are calculated and scaled to the same range of
values, i.e. the same minima and maxima like the first spectral range.
Example: The spectral distances in the first spectral range are between 2
and 10, in the second between 6 and 22. The distances of the second
spectral range are transformed as follows:
D → 0.5 ⋅ D – 1
(6-5)
After this transformation the spectral distances of the second spectral
range have the same values as the distances belonging to the first spectral range.
As the scaling has referred to the first spectral range, it does matter
which spectral range is selected first. Make sure to select the correct
spectral range as first range.
If spectral distances are sorted according to ascending values, this order
directly results from the spectral ranges selected. For example, when
comparing a test spectrum with a reference spectrum using an IDENT
60
OPUS/IDENT
Bruker Optik GmbH
Data Preprocessing
test, the spectral distance may have the lowest value to the fourth reference spectrum (best Hit Quality) based on the first spectral range. If you
consider the second spectral range, the spectral distance may have the
lowest value to the eleventh reference spectrum.
4b) When using the Normalize to Reprolevel method you have to define a
reproduction level (see figure 71) for each spectral range. The spectral
distances will be divided by this reproduction level. This is the reason
why the spectral distances are indicated as reproduction level units,
which means that you can set a threshold for the identity test. For example, if the Hit Quality is below 1 in case of a test spectrum, the sample is
regarded as being Identified. If, however, the spectral distance is above
1, the spectrum cannot be assigned to any reference spectrum.
5) Irrespective of the method used, spectral distances can be weighted for
each single spectral range (see chapter 7.3), according to the following
equation:
∑ wj ⋅ Dj
D = ---------------------(6-6)
w
∑ j
Spectral distances calculated by the Normalize to Reprolevel algorithm
may be above 2000, if reproduction levels are other than 1. When using
the Scaling to 1st Range method, the values have to be between 0 and
2000.
6.3
Data Preprocessing
OPUS provides several data preprocessing methods.
6.3.1
Vector Normalization
The maximum value of the Hit Quality has to be defined only if Vector
Normalization was used to preprocess data. If you use a preprocessing method
other than Vector Normalization, no upper limit for the Hit Quality has to be
defined, i.e. you can use any numerical value. The maximum spectral distance
is 2 (maximum difference of the spectra) in case of Vector Normalization,
provided you have selected Standard method.
Vector Normalization first calculates the average y value of spectra and only
uses data points within the selected spectral ranges. The average value
calculated will then be subtracted from the spectrum, which causes the spectrum
to be centered at around y = 0. This is followed by calculating the sum of
squares of all y values, and the respective spectrum is divided by the square root
of this sum. The vector norm of the result spectrum is 1:
am
Bruker Optik GmbH
∑
a(k)
= -------------------N
k
OPUS/IDENT
(6-7)
61
IDENT Theory
a′ ( k ) = a ( k ) – a m
(6-8)
a′ ( k ) a″ ( k ) = -------------------------------∑ ( a′ ( k ) )2
(6-9)
k
∑ ( a″ ( k ) )2 =
k
1
(6-10)
If vector normalized spectra are represented in n dimensional space and n being
the number of selected data points, all spectra are on the unit sphere (n
dimensional sphere around the coordinate origin with radius 1, see figure 46).
The maximum distance between two spectra is the diameter of the unit sphere,
i.e. 2.
r=1
D
Figure 46: Two vector-normalized spectra on the unit sphere
To explain this in more detail, create a new spectrum. Invert one reference
spectrum from the example library, i.e. multiply the spectrum by -1 using the
OPUS Spectrum Calculator.
Compare the inverted spectrum with the reference spectra. Select an IDENT
method that preprocesses data by Vector Normalization. Figure 47 shows the
original spectrum (top) and the inverted spectrum (down). Figure 48 shows the
identity test result. The last hit in the result list is the original reference spectrum
with a spectral distance of 2 compared to the test spectrum.
62
OPUS/IDENT
Bruker Optik GmbH
Data Preprocessing
Figure 47: Original and inverted spectrum
One advantage of Vector Normalization is that the range of values for Hit
Quality is known (from 0 to 2). This simplifies the interpretation of the identity
test result. Additionally, using Vector Normalization as data preprocessing
provides an even more important aspect.
Bruker Optik GmbH
OPUS/IDENT
63
IDENT Theory
Figure 48: Ident Result of inverted spectrum searching
Vector Normalization also reduces the differences between each single
measurement of the same sample. Figure 49 shows 11 spectra acquired from
one single sample. As the substance has been powder the single spectra differ
substantially from each other. These differences can be considerably reduced by
using Vector Normalization.
Note the different scaling of the ordinate (y axis) in figure 49. Spectra derived
from the same sample have to show only very small differences. Therefore,
Vector Normalization is highly recommended in these cases. Further data
preprocessing methods are First and 2nd Derivative.
64
OPUS/IDENT
Bruker Optik GmbH
Data Preprocessing
Figure 49: Original and vector-normalized spectra
Bruker Optik GmbH
OPUS/IDENT
65
IDENT Theory
Repeat the identity test for the first analytical example and select Vector Normalization to preprocess data. Figure 50 shows the result of the new analysis.
Compare these results with the ones shown in figure 40.
Figure 50: Ident Report – example of vector normalization
The lowest spectral distance is 0.009, the next (0.33) is higher, i.e. by about a
factor of 30. This factor is higher than the one obtained without using Vector
Normalization (factor 6).
The reference spectrum of Hit No. 1 is L-Leucin. This identification is correct
as the test spectrum has been measured from the same sample. In general,
however, the question is how far the spectral distance (Hit Quality) may
increase to be still within an acceptable threshold to correctly identify the test
spectrum?
To define such a threshold, it is not sufficient to measure only one single
spectrum per reference substance. You must measure several spectra and
determine this threshold from spectral fluctuations (spectral differences). The
66
OPUS/IDENT
Bruker Optik GmbH
Determining Threshold Value for Identity Test
average spectrum calculated from each measurement is then added to the
IDENT library as reference spectrum.
6.4
Determining Threshold Value for
Identity Test
There are two possibilities to define the limit value for an IDENT group:
1) Fixed Algorithm (Maximum Hit + x SDev.)
The threshold is calculated from the worst hit (i.e. the largest Hit Quality
value in the average report) and the standard deviation S0:
Threshold D T = D Max + S 0 ⋅ x
(6-11)
whereas the default x value is 0.25.
The threshold is selected so that all original spectra used to create the
reference spectrum (average spectrum) have a lower distance than this
threshold to the reference spectrum.
If the analysis of a sample spectrum produces a spectral distance which
is larger than this threshold, the sample spectrum will be defined as not
being identical.
2) Confidence Level
Two parameters are derived from the spectral distances (see above) to
define the confidence region for the average spectrum:
The mean distance DM
DM =
( i )∑ D---------n
(6-12)
i
The standard deviation S0
S0 =
∑ D ( i )2
i
--------------------n–1
(6-13)
with n being the number of original spectra. Note that S0 is the standard
deviation from zero and not the standard deviation SM from the mean
value.
The standard deviation from the mean distance SM can be calculated
from DM and S0:
SM =
Bruker Optik GmbH
∑i ( D ( i ) – DM )
2
----------------------------------------- =
n–1
OPUS/IDENT
2
S0 –
2
DM ⋅ n
--------------n–1
(6-14)
67
IDENT Theory
The threshold is calculated by multiplying the standard deviation SM
with a factor f and adding the mean distance DM:
DT = DM + f ⋅ SM
(6-15)
The factor f is calculated from the probability Φ which can be chosen
between 95% and 99.9999%. Example: if you choose Φ = 97.7%, factor
f will be 2. The spectral distances are assumed to be distributed according to a normal distribution. Note that the probability value is calculated
for a single-sided limit.
If you select Φ = 95%, then 5% of the original spectra are outside the
confidence region (the spectral distance to the average spectrum is larger
than the threshold in question). Whether this is actually correct or not
can be tested by a validation. For details see chapter 6.7.
3) Abs. Threshold
This option allows you to define the threshold for each reference spectrum. This threshold, however, will only take effect if the group consists
of several spectra. If you define the threshold, make sure to consider the
spectral differences of the original spectra.
6.5
Identity Test
The identity test routine generates a hit list which is stored in the IDENT report,
sorted by ascending spectral distances. Provided the expected reference has
been selected automatically from sample name, or has been defined by the user
the following three categories of identity test results are possible: Identical, Not
Identical or Can Be Confused With (see below).
If a group consists of only one spectrum, the identity test result will be: Identity
Not Checked. If no expected reference has been defined, the identity test results
will be: Identified As, Not Identified, No Unique Identification Possible.
Identical to (in case of expected reference):
The first hit must be the expected reference. The Hit Quality of the first
substance (i.e. the spectral distance between a test spectrum and the first
reference spectrum in the report) has to be smaller than the threshold of the
expected reference, and the spectral distances between the query spectrum and
all other average spectra are always larger than the corresponding thresholds.
Identified As (in case of no reference defined):
All threshold values are taken into account. In case of Identified As, the Hit
Quality of the first substance (i.e. the spectral distance between a test spectrum
and the first reference spectrum in the report) has to be smaller than the
corresponding threshold. But the Hit Quality of all other substances has to be
larger than the corresponding threshold.
68
OPUS/IDENT
Bruker Optik GmbH
Identity Test
Not Identical to (in case of expected reference):
The spectral distance between the test spectrum and expected reference
spectrum is larger than the threshold value.
Not Identified (in case of no reference defined):
In this case the Hit Quality of all reference substances has to be larger than the
threshold.
Can be Confused with (in case of expected reference):
This result indicates that the spectral distance of the test spectrum to at least one
other average spectrum is smaller than the confidence region. If, e.g., the Hit
Quality of 4 substances is smaller than the corresponding threshold and one of
these substances is identified to be the expected reference.
No Unique Identification Possible (in case of no reference defined):
The Hit Quality of more than one substance is smaller than the corresponding
threshold. Code numbers are assigned to the individual test results. Therefore,
the results can be easily evaluated in an OPUS macro.
Can be confused with <N> Other Hits (in case of expected reference)
The code number 2 is used if all reference spectra with a Hit Quality less than
the threshold have the same Sample Name (or Sub Sample Name) as the query
spectrum. The value of 2 may only occur if the Selected automatically from
Sample Name option button on the Expected Reference dialog (figure 87,
page 104) has been activated.
Bruker Optik GmbH
OPUS/IDENT
69
IDENT Theory
Table 2: Code Numbers of Identity Test
Identity Test Result
Code
Identity not Checked (no threshold available)
0
Identical To/Identified As
1
Can be Confused with/No Unique Identification Possible
-1
Not Identical/Not Identified
-2
Can be confused with <N> Other Hits
2
ER
ER
D
DT
D
TS
DT
TS
D < DT
D > DT
Identified
Not Identified
Figure 51: Schematic representation of Identity Test Results Test Spectrum (TS), Expected Reference (ER)
6.6
Class Test
Ideally, all reference spectra are well-distinguishable from each other and the
thresholds are so small that their confidence regions do not overlap. This
situation is shown in figure 52. The reference spectra are dots (in a
mathematical sense) in the n dimensional space, n being the number of data
points selected. The confidence regions can be depicted by spheres which
centers represent the reference spectra.
70
OPUS/IDENT
Bruker Optik GmbH
Class Test
Figure 52: Library with well-distinguishable reference spectra
However, it may occur that the confidence regions of some reference spectra do
overlap, see figure 53. As you can see, the confidence regions of the A, B, and C
reference spectra clearly overlap.
B
A
C
Figure 53: Three reference spectra with overlapping confidence regions
It is possible to define several reference spectra which are members of one
class, and to perform a class test during the IDENT analysis. Both in case of an
expected reference as well as in case of an analysis with no reference defined,
the class test determines whether all reference spectra with their Hit Quality
below the corresponding threshold are members of the same class. If so, the
result will be Class Test OK. Otherwise, the result will be Class Test NOT OK.
If the expected reference spectrum is not part of any class, the IDENT report
says Class Test NOT PERFORMED.
Bruker Optik GmbH
OPUS/IDENT
71
IDENT Theory
Make sure to define only one class name for one group in the IDENT method. If
you load a previously used method including extraordinary members, only the
first class name will be considered.
All class members use their individual thresholds. If you load a previously used
method when using the Setup Identity Test Method command from the Evaluate
menu, the exclamation mark (!) next to the old class name is automatically
deleted.
Figure 54: IDENT Report with class test performed - Test OK
72
OPUS/IDENT
Bruker Optik GmbH
Class Test
Figure 55: IDENT Report with class test performed - Test Not OK
Code numbers are assigned to the individual class test results. This causes the
results to be easily evaluated in an OPUS macro.
Table 3: Code Numbers of Class Test
Class Test Result
Code
Class Test OK
1
Class Test not Performed
0
Class Test not OK
-2
Bruker Optik GmbH
OPUS/IDENT
73
IDENT Theory
6.7
Validation
When setting up an IDENT library, you have to check whether the IDENT
method parameters are optimized for all reference spectra in one library. This
can be done by Validation, which compares original spectra with average
spectra.
Figure 56 shows a validation report, using the Standard method. All original
spectra have been compared with the average spectra of the library. The
Abs. Threshold values are used as confidence region in the validation process.
The results are either Uniquely Identified, Not Identified and Can Be Confused
With, similar to the Identity Test (see chapter 6.5). The number of spectra which
belong to the respective class is indicated at the end of the validation report. The
total of all spectra has to be equal to the total of the original spectra tested.
In case of overlappings the Detailed and Result reports provide additional
information on which groups should be assigned to a new common sub-library.
Figure 56: Validation Report showing assignment recommendation
An original spectrum is Uniquely Identified if the spectral distances between
this spectrum and the corresponding average spectrum is smaller than the
threshold, while the spectral distances between the original spectrum and all
other average spectra are larger than the corresponding confidence region. The
original spectra which are Uniquely Identified are not listed in the report.
74
OPUS/IDENT
Bruker Optik GmbH
Validation
If a spectrum is Not Identified, the spectral distance between the original
spectrum and average spectrum is larger than the threshold. In this case the
report indicates the corresponding average spectrum, sample name and
threshold (confidence region) specified for this substance. The file names of the
original spectra which have not been identified, and the spectral distances
between these original spectra and the average spectrum (Hit) are listed under
the Original Spectra Outside Confidence Region definition.
Figure 57: Validation Report – Spectra identified
The Can Be Confused With result indicates that the spectral distance of an
original spectrum to the corresponding average spectrum is smaller than the
confidence region, while one or more spectral distances between the original
spectrum and other average spectra are even smaller than the corresponding
confidence regions.
If an original spectrum is tested to be Can Be Confused With other references,
first, its average spectrum, sample name and confidence region are listed in the
report, followed by the name of the original spectrum and the threshold. In
addition, the name of the average spectrum, the sample name and the spectral
distance (Hit) between this average spectrum and the original spectrum are
listed under the Overlapping With definition. The spectral distance is smaller
than the threshold (confidence region). If this original spectrum overlaps several
average spectra, they will all be listed in the report.
Bruker Optik GmbH
OPUS/IDENT
75
IDENT Theory
Figure 58: Validation Report – spectra can be confused with
If all reference spectra are selected, each original spectrum will be compared
with all average spectra. Even if you only select some of the reference spectra,
the original spectra belonging to these reference spectra are tested against all
average spectra. This option is extremely useful if an existing library is to be
extended by new reference spectra and only these new spectra have to be tested.
If you have activated the Always use lowest IP level check box on the
Parameters tab, the detailed report will only list the results of the lowest IP
level for all spectra of each group.
76
OPUS/IDENT
Bruker Optik GmbH
Setup Identity Test Method - Load Method
7
Reference Section
Before being able to identify spectra by means of OPUS IDENT you have to
create an IDENT method first. Select the Setup Identity Test Method command
from the OPUS Evaluate menu.
7.1
Setup Identity Test Method Load Method
Figure 59: Setup Identity Test Method – Load Method tab
Use the Load Method button to load an existing IDENT method. IDENT
method files have the extension *.FAA. It is also possible to load IDENT
method files created by OPUS-OS/2. However, if you store such an OS/2
method using OPUS IDENT, you will not be able to load the method by OPUSOS/2 IDENT again. To solve this problem, store the modified OPUS-OS/2
IDENT file by using a different file name.
The General information of selected library group field provides statistical
information on the existing method file. The number of spectra used for the
method, and the number of frequency ranges included are displayed.
Bruker Optik GmbH
OPUS/IDENT
77
Reference Section
You will get additional information on the data preprocessing method, the
algorithm used for the identity test, and how many sub-libraries are part of the
reference library.
For further details on how to rename libraries, refer to chapter 1. To print the
library structure use the drop-down list to select the respective printing option
and click on the Print button. For further details see chapter 1.
7.2
Setup Identity Test Method Reference Spectra
Figure 60: Setup Identity Test Method – Reference Spectra tab
The spectra table lists the spectra groups and each single spectra, including the
sample ID, Path, File Name, Sample Name, Group Name and Sub Library. You
can have the groups displayed as well as each single spectra of one group. Click
on the
sign in the first column. The respective line with the group selected
opens, and shows the single spectra. To close the list again, click on the sign.
Click on the numbered tiles on the left side of the table to select one spectrum or
several spectra. Select the whole table by clicking on the tile on the left side of
the table header (see mark in figure 60). Remove spectra from the table by
selecting one spectrum or more spectra and pressing the DEL key on the PC
keyboard.
78
OPUS/IDENT
Bruker Optik GmbH
Setup Identity Test Method - Reference Spectra
If you click on the Add Spec. for New Group button, a dialog box opens to be
used to load one or more files into the spectra list. Spectra loaded
simultaneously will be merged into one group. Use the Add Spec. to Sel. Group
button to add a spectrum to a group selected. This is useful if you want to add a
spectrum to a group later.
7.2.1
Sorting Reference Spectra
When working with IDENT methods consisting of a large number of groups it
takes quite some time to find a particular group. Therefore, you can sort the
reference spectra by ID, sample or group name and sub-library in ascending or
descending order. Double click on the respective column. Note that if you sort
the spectra, e.g., by sample name, the ID number in the first column, however,
will keep the original order (see figure 61).
Figure 61: Setup Identity Test Method – Reference spectra sorted in descending order
7.2.2
Missing Reference Spectra
If you load an IDENT method, it may occur that certain spectrum files listed in
the particular method are missing in the data path. To be able to localize the
missing spectra the group name which the missing spectra belong to will be
highlighted in red. If you click on the
sign of the particular group, the name
of the missing spectra will also be highlighted in red.
Bruker Optik GmbH
OPUS/IDENT
79
Reference Section
Figure 62: Setup Identity Test Method – Missing reference spectra
Missing reference spectra may be due to file renaming or, as exemplified in
figure 62, to a different data path. In this case store the missing reference
spectra into the right path and load the IDENT method again. This is important
as you cannot perform any calculation using an IDENT method with missing
spectra. If you want to know the total number of spectra included in a particular
group, click on the Threshold tab. The Spectra column includes the total
number of spectra per group.
Figure 63: Total number of spectra within one group
80
OPUS/IDENT
Bruker Optik GmbH
Setup Identity Test Method - Reference Spectra
7.2.3
Options
It is also possible to add average spectra generated by previous OPUS IDENT
versions without requiring to reconstruct the original spectra which the average
has been generated from. Each average spectrum loaded represents one group.
To switch between original and average spectra generated by previous OPUS
versions, first click on the Options button. The following dialog opens:
Figure 64: Options
The Add selected spectra into one group option button is activated by default. In
this case all spectra commonly selected during loading are added to one group.
In the Options dialog box you define the paths for the original and average
spectra. Additionally, you can define the group name which is derived from the
respective sample name. Enter the position of the first character as well as the
length of the sample name.
Note: Once defined, the group name should not be changed any more to avoid
confusion.
7.2.4
Set Sub Library
The Set Sub Library button on the Reference Spectra tab enables you to add
sub-libraries to the current main method. Select the New option from the dropdown list to enter a unique sub-library name for the spectra groups defined.
Choose the group(s) which have to be assigned to this new sub-library from the
Select groups for... selection field. There are only groups available which have
not yet been assigned to another sub-library on the same library level. Click on
the Assign button.
Bruker Optik GmbH
OPUS/IDENT
81
Reference Section
To delete sub-libraries select them first and click on the Delete Sub Library
button.
Figure 65: Set Sub Library
82
OPUS/IDENT
Bruker Optik GmbH
Setup Identity Test Method - Reference Spectra
7.2.5
Assign Classes
To assign classes to an existing method click on the Assign Classes button. The
following dialog opens:
Figure 66: Assign Classes
Select the New option from the drop-down list to enter a unique class name for
the spectra groups defined. Choose the group(s) which have to be assigned to
this new class from the Select groups for... selection field. There are only
groups available which have not yet been assigned to another class on the same
library level. Click on the Assign button.
To delete classes select them first and click on the Delete Class button.
Bruker Optik GmbH
OPUS/IDENT
83
Reference Section
7.3
Setup Identity Test Method Parameters
Figure 67: Setup Identity Test Method – Parameters tab
7.3.1
Preprocessing
You can select several data processing methods from the drop-down list: Vector
Normalization, First and 2nd Derivative as well as combinations of both
methods.
• Vector Normalization
The Vector Normalization data preprocessing normalizes a spectrum,
i.e. the average y value is calculated first and subsequently subtracted
from the spectrum. Then, the sum of squares of all y values is
calculated and the spectrum is divided by the square root of this sum.
This method is used in case of different optical thickness to compare
the samples with each other. The form of the different spectra will be
preserved, which facilitates the interpreting of spectra. However, the
result extremely depends on the spectral region selected, i.e. specific
differences of one region are distributed to all data points.
84
OPUS/IDENT
Bruker Optik GmbH
Setup Identity Test Method - Parameters
• First Derivative
Calculates the first derivative of the spectrum by interpolation. Steep
edges of a peak become more important compared to flat structures.
This method is mainly used to preprocess pronounced, but small
features which are overlaid by a high and broad background.
In case of this method the window size selected is very important.
The smaller the window size, the more spectral details are shown,
with the spectral-to-noise ratio being apparently higher.
• 2nd Derivative
This method is similar to First Derivative, but it allows to evaluate
extremely flat structures.
If you use one of the derivative methods, an additional selection field will be
displayed to define the amount of smoothing points. You can select between 5
and 25 points. The optimal number of smoothing points, however, has to be
evaluated empirically.
Figure 68: Defining smoothing points
7.3.2
Regions
The Regions table allows to limit data to one or several spectral regions to be
considered for identification. The frequency limits for the spectral regions can
either be entered manually or selected interactively.
7.3.3
Interactive Region Selection
The spectral region shown on the white background will be processed and
evaluated. You can also modify the spectral regions displayed. Place the cursor
on the boundary between the gray and white area. Press the left mouse button
and move the regions. It is also possible to move the entire spectral region. If
you position the cursor on the white area, the cursor changes from
into
.
Press the left mouse button and move the spectral region. To delete a region,
right click on the white area and select Remove from the pop-up menu.
Bruker Optik GmbH
OPUS/IDENT
85
Reference Section
Figure 69: Select Frequency Range(s)
If you click on the Interactive Region Selection button, a separate window
opens and displays the reference spectra. You can add a new spectral region by
right-clicking on the window and selecting the Add Region option from the popup menu. The pop-up menu also includes the Zoom and Crosshair option. These
options allow to easily define the value of a specific data point.
Figure 70: Select Frequency Range(s) with pop-up menu
86
OPUS/IDENT
Bruker Optik GmbH
Setup Identity Test Method - Parameters
7.3.4
Clear Selected Regions
You can delete an entry from the Regions table by selecting the specific entry
and clicking on the Clear Selected Regions button or pressing the DEL button
on your keyboard.
7.3.5
Method
Use this drop-down list to select an identification method. During the identity
test the test spectrum is compared with all reference spectra. The result of this
comparison is the spectral distance which is also called Hit Quality. The more
similar two spectra are, the smaller the spectral distance. Four methods are
available to calculate the spectral distance. Use the Method drop-down list to
select one. The basic theory of each method has been described in chapter 6.
To enforce an IDENT analysis on the lowest IP level during the identity test the
Always use lowest IP level check box has to be activated. This check box is only
enabled on the very first library level as this a global setting for the entire
library structure, and is deactivated by default. See also chapter 1.
IDENT methods which have been stored created and stored by previous OPUS
without this algorithm can be loaded as well to perform this kind of analysis.
Depending on the method selected, additional columns are added to the Regions
table, e.g. Weight and Reprolevel.
Figure 71: Regions list with weight and reprolevel columns
Bruker Optik GmbH
OPUS/IDENT
87
Reference Section
You can select between the following identification methods:
• Standard
The Standard method calculates the Euclidean distance between the
test and reference spectra.
• Factorization
The factorization is performed on average spectra of the respective
groups. The spectra are first represented as linear combination of the
factor spectra and the resulting coefficients are used to calculate the
spectral distance.
• Factorization (orig. specs)
The factorization is performed on all original spectra of the
respective groups.
• Scaling to 1st range
Performs the Scaling to 1st range algorithm. For details, see
section 6.2.1.
• Normalize to Reprolevel
Performs the Normalize to Reprolevel algorithm. For details, see
section 6.2.1.
7.3.6
Calculate Thresholds
If you click on the Start Calculation button, you start the calculation. If you
select the Factorization or Factorization (orig. specs.) algorithm the additional
Set Factors button will be displayed. Click on the Set Factors button to open the
following dialog:
Figure 72: Factor Spectra
88
OPUS/IDENT
Bruker Optik GmbH
Setup Identity Test Method - Parameters
Selecting the optimum number of factors is not easy. OPUS facilitates this
procedure by highlighting those factors which consists of the most nonoverlapping score coefficient clusters, as shown in figure 72. The following
graph is a simplified depiction of such non-overlapping score coefficient
clusters of two groups A and B:
A
B
Score coefficients
Example:
The factor selection field in figure 72 has to be interpreted as follows:
A
B
C
Figure 73: Factor spectra highlighted
A)
B)
C)
Eigen values
Number of separated score coefficient clusters (it is always started with the factor
which includes the most groups separated)
Number of separated score coefficient clusters additional to the factors selected before.
In case of factor 6, e.g., there are 120 clusters separated, with 10 of these 120 clusters
not being separated by factor 5. In case of factor 7 there are 118 clusters separated,
with 3 of these 118 clusters not being separated by factor 5 nor factor 6.
Select the factor spectra from the list, which are used to calculate the spectral
distance. It is not advisable to accept this value without performing a validation
first. We recommend to perform several validations of the IDENT library using
different numbers of factor spectra. Select the optimum number of factor
spectra according to the result obtained.
It is not necessary to use a consecutive sequence of factor spectra. You can
select factor 2, 3 and 5. Delete factor 1 if you do not want to get any baseline
information.
If you want to calculate spectral distances by using spectral residuals, activate
the Use Residuals check box.
Bruker Optik GmbH
OPUS/IDENT
89
Reference Section
7.4
Setup Identity Test Method – Threshold
Figure 74: Setup Identity Test Method – Threshold tab
The threshold of a reference spectrum is the sum calculated by the maximum
distance (maximum Hit Quality) listed in Group Statistics, plus the amount
resulting from standard deviation (SDev) and a user-defined x factor. The
threshold values, listed in the Threshold column are indicated for each reference
spectrum.
7.4.1
Maximum Hit + X*SDev
This formula is used to calculate the threshold value. You can enter any value
for X in the entry field, with 0.25 being set as default. To confirm your entry
click on the Set button. This causes the new value to be set for all reference
spectra, and the Threshold values will be updated.
7.4.2
Confidence Level
Two parameters are derived from the spectral distance to define the confidence
level. You can enter any factor between 95 and 99.9999% into the entry field,
99.99% is set by default. To confirm your entry click on the Set button. See also
chapter 1.4.
90
OPUS/IDENT
Bruker Optik GmbH
Setup Identity Test Method – Threshold
7.4.3
Set
If you click on the Set button, all changes made will be displayed on the table
list. If you have selected only part of the column, only the values of the lines
selected will be changed.
7.4.4
Group Statistics
The Group Statistics parameter includes detailed information on the Group and
File Name, Hit Quality, Standard Deviation and Mean Distance.
Two parameters are derived from the spectral distances between original spectra
and the average spectrum to define the confidence region for a group:
DM mean distance:
DM =
( i )∑ D---------n
(7-1)
i
S0 standard deviation:
S0 =
∑ D ( i )2
i
--------------------n–1
(7-2)
with n being the number of original spectra.
Select the Group Statistics option from the selection box. The following dialog
opens:
Bruker Optik GmbH
OPUS/IDENT
91
Reference Section
Figure 75: Setup Identity Test Method - Group statistics
If you use the confidence level to calculate the threshold, the Outlier column is
very helpful to see at once the number of spectra per group, which are outside
the threshold. To have the Hit Quality and File Name displayed click on the
button of the respective Group Name line.
92
OPUS/IDENT
Bruker Optik GmbH
Setup Identity Test Method – Validate
7.5
Setup Identity Test Method – Validate
Figure 76: Setup Identity Test Method – Validate tab
Start validation by clicking on the Validate button. The following menu pops
up:
Figure 77: Validate pop-up menu
Bruker Optik GmbH
OPUS/IDENT
93
Reference Section
Two validation options are available:
• Validate this library
This option validates the main and sub-library you are currently
working with.
• Validate this library and all sub-libraries below
This option validates the main and sub-library you are currently
working with, and all additional sub-libraries belonging to the
current library.
The progress of the validation process is indicated by the status bar.
7.5.1
Validation Report
Validation reports will not be overwritten and the report file name is always the
same defined for the main or sub-library. Groups assembled in common classes
are not considered to overlap. Nevertheless, the IDENT reports include the
groups assigned to classes. You can select between the following reports which
are based on single spectra, except for the selectivity report and histogram:
• Summary Report
The Summary Report outlines all important information on the
current IDENT method, e.g. path and file name, sub-library names,
date, time, operator name and comments. It includes all the groups
which can be confused with other groups.
• Result Report
The Result Report outlines all important information on the current
IDENT method, e.g. path and file name, sub-library names, date,
time, operator name and comments. It gives additional information
on which groups should be assigned to a new common sub-library.
• Detailed Report
The Detailed Report includes additional information on the
algorithm used and frequency ranges defined of all sub-libraries, the
order of internal derivative and smoothing points for internal
derivative. Furthermore, this reports specifies all thresholds of the
overlapping groups as well as the distances of all single spectra
which overlap. The report provides additional information on which
groups should be assigned to a new common sub-library.
If you have activated the Always use lowest IP level check box on the
Parameters tab, the Detailed Report will include only the results of
the lowest IP level for all spectra of each group.
• Selectivity Report
This report is based on average spectra. If you use the selectivity
slider on top, you can get a more detailed report. For example, if you
set the slider to a selectivity of 3, all spectra are shown in the spectral
distance between 1 and 3.
94
OPUS/IDENT
Bruker Optik GmbH
Setup Identity Test Method – Validate
• Selectivity Histogram
The histogram is a summary of the selectivity report.
In general, validation reports directly compare one spectra group with the
adjacent spectrum to see which clusters overlap and where. The selectivity
report compares Material 1 and Material 2, as explained in the chart in
figure 78.
Figure 78: Calculating selectivity
D
The selectivity will be calculated as follows: S = ---------------------with S being the
( T1 + T2 )
ratio of distance D between average spectra and the sum of threshold values T1
and T2 (cluster radii). This results in the following:
• S < 1: overlapping
• S = 1: cluster in contact
• S > 1: cluster separated
Figure 79 exemplifies a selectivity report.
Bruker Optik GmbH
OPUS/IDENT
95
Reference Section
Figure 79: Selectivity Report
The selectivity report displays the result in different colors:
•
•
•
•
Red: spectra with a selectivity of <1
Green: spectra with a selectivity of >2
Black: spectra with a selectivity between 1 and 2
Gray: spectra with a selectivity of <1, without single-spectra
overlappings in the validation report
Note: Generally, single spectra are not relevant in case of selectivity, however, they
are indicated in the validation report. If there are no single-spectra overlappings in
this report, the respective group is displayed in gray in the selectivity report.
It may occur that in case of libraries with reference spectra not any of these
spectra is within the intersection of two clusters. Therefore, the validation
would yield to a non-overlapping result. However, the selectivity does indicate
the geometric overlapping, as exemplified in figure 80.
96
OPUS/IDENT
Bruker Optik GmbH
Setup Identity Test Method – Validate
Figure 80: Selectivity - Geometric overlapping
The selectivity report can also be read out as score plots in 3D-format, which is
indicated by the
tab. This kind of representation is based on the
factorization of single reference spectra or average spectra, shows the
distribution of spectra and supports the selecting of meaningful factor spectra.
Select Factorization from the Method drop-down list on the Parameters tab and
click on the Start Calculation button. Define at least 3 factors from the Factor
Spectra dialog which will serve as a basis for the 3D factor view. Subsequently,
validate the library by clicking on the Validate button on the Validate tab.
To display the cluster of one group select the respective group from the
Selectivity Report list. The number of neighboring groups can be set by means
of the Selectivity slider. You can activate or deactivate the Opaque check box
(see figure 79). In both cases the real threshold of the cluster is shown in all
dimensions, i.e. x, y and z axis. If you deactivate the Opaque check box, the
spectra (A in figure 81) of each cluster can be seen as the clusters will be
transparent.
Bruker Optik GmbH
OPUS/IDENT
97
Reference Section
A
Figure 81: 3D factor view - Transparent clouds and data points
If you position the mouse on one specific spectra, the file name and group name
will be displayed. To improve the factor view, you can rotate the box. If you
position the mouse on the edge of the box, the cursor changes into
. To
rotate the box press the left mouse button and move the mouse to the position
desired.
Right clicking somewhere on the 3D display pops up the Properties button. If
you click on this button, the View properties dialog is displayed which allows
further plot settings. To optimize the 3D factor view you can select additional
factors by means of the Factor drop-down list. Always the next 3 factors of the
original selection can be used for each dimension.
98
OPUS/IDENT
Bruker Optik GmbH
Setup Identity Test Method – Validate
Figure 82: View properties
Sometimes the clusters may be displayed as ellipsoides. This is due to the
scaling of the axes, and not a result of the original calculation.
To zoom the box displayed press the left mouse button and draw a frame around
the position desired. Double click into the zoomed area to undo the zoom
setting. To return to the IDENT setup close the factor view by clicking on the
icon.
7.5.2
Print
To print the report, click on the Print button. This starts the Windows Notepad
program which you can also use to reformat the text, if desired. Use the
Notepad print function to create a printout.
Note: It is recommended to select a small font in Windows Notepad to avoid
extraordinary long reports. A proportional font may lead to a confusing display of
the results. Therefore, it is advisable to use a monospace font, e.g. Courier New,
10.
Bruker Optik GmbH
OPUS/IDENT
99
Reference Section
7.6
Setup Identity Test Method Store Method
Figure 83: Setup Identity Test Method – Store Method tab
This dialog allows to store a method file created. The parameter you can define
is the number of Hits to be stored in the IDENT report. By default, the Store
Average Spectra is activated. Click on the Store Method button to open the
standard Save File dialog box. The method file has the extension *.FAA. All
sub-libraries will be stored simultaneously.
100
OPUS/IDENT
Bruker Optik GmbH
Identity Test
7.7
Identity Test
To start an IDENT analysis select the Identity Test command from the Evaluate
menu. The following dialog box opens:
Figure 84: Identity Test - Select File(s) tab
Select a test spectrum and drag and drop the spectrum absorption block from the
OPUS browser window into the File(s) for identity test selection field.
To load or modify an IDENT method click on the Load Ident Method button
and select the desired method from the dialog that opens. If an identity test
method has already been loaded (e.g. if you have created one prior to starting
the analysis) the path and method name will be indicated in the Loaded Identity
Test Method field.
The Show results immediately check box in the Output options group field is
activated by default and the identity test results will be shown in a special
evaluation result display, see figure 85 and 86.
Bruker Optik GmbH
OPUS/IDENT
101
Reference Section
Figure 85: IDENT Evaluation Result display - Result OK
In the lower part of the display the Identity Test result is indicated. A green
check mark and the description OK would indicate that the test has passed, i.e.
the product has been identified.
A red cross and the description NOT OK would indicate that the comparison has
failed, i.e. the product has not been identified.
102
OPUS/IDENT
Bruker Optik GmbH
Identity Test
Figure 86: IDENT Evaluation Result display - Result not OK
To print the evaluation result display of the identity test, activate the Print
Results Automatically check box in the Output Options group field.
7.7.1
No Reference Defined
If there is no expected reference defined, click on the Change button. There are
three possibilities to define an expected reference:
Bruker Optik GmbH
OPUS/IDENT
103
Reference Section
Figure 87: Expected Reference
a) No reference defined
The identity test analysis works without a pre-defined reference
spectrum.
b) Selected automatically from sample name
The reference spectrum is determined by comparing the sample name
of the test spectrum with the sample names of the library spectra.
Usually, the sample name will not be completely used, but partially.
First character:
Indicates at which sample name character the character comparison will
start. In this example the comparison starts at character 1.
Length:
The Length indicates how many characters will be taken into account
during comparison. Example: the sample name of the test spectrum is
000002, Sample DL-Isoleucin. It is sufficient to use the first six characters for a definite selection: First = 1 and Length = 6.)
c) User-selected from table below
If you activate this option button, all reference spectra will be listed.
Each line contains the sample name (e.g. 000002, Sample DLIsoleucin), the number of the reference spectrum in the library (e.g.
ID=2) and the group name. Select the spectrum which you expect to
match the test spectrum. The test spectrum can have any sample
name.
104
OPUS/IDENT
Bruker Optik GmbH
Cluster Analysis – Load Method
In the Sort group field you define how to sort the list. You either sort according
to Sample name, ID or Group Name. If you check Sample name, you can
additionally define the character number of the sample name, which you start
sorting with.
7.8
Cluster Analysis – Load Method
Figure 88: Cluster Analysis – Load Method tab
7.8.1
Load Method
Use the Load Method button to load an existing cluster analysis method. Cluster
analysis method files have the extension *.CLA. It is also possible to load
cluster analysis method files created by OPUS-OS/2 IDENT. However, if you
store such a method using OPUS/IDENT, you will not be able to load the
method by OPUS-OS/2 IDENT. To avoid this, store the modified OPUS-OS/2
IDENT file by using a different file name.
Bruker Optik GmbH
OPUS/IDENT
105
Reference Section
7.8.2
General Information
The General information group field provides statistical information on the
existing method file. The number of spectra used for the method and the number
of frequency ranges included are displayed. You will get additional information
on the data preprocessing method, the algorithm used for the identity test and
dendrogram, and whether a distance matrix has been generated.
7.9
Cluster Analysis – Reference Spectra
Figure 89: Cluster Analysis – Reference Spectra tab
This dialog box is the same as the Reference Spectra dialog box of the Setup
Identity Test Method command and has been described in section 7.2.
106
OPUS/IDENT
Bruker Optik GmbH
Cluster Analysis – Parameters
7.10
Cluster Analysis – Parameters
Figure 90: Cluster Analysis – Parameters tab
7.10.1
Preprocessing
The cluster analysis uses the same data preprocessing methods as described in
section 7.3.
7.10.2
Regions
The Regions table allows to limit the data to one or several spectral regions to
be considered for the cluster analysis. The frequency limits for the spectral
regions can either be entered manually or selected interactively.
Bruker Optik GmbH
OPUS/IDENT
107
Reference Section
7.10.3
Method
Select an algorithm to identify the spectrum (see section 6.1 and 7.3).
If you select the Factorization method, you have to specify the number of factor
spectra to be used to calculate the spectral distances, in the Factor Spectra
dialog box. This dialog box opens automatically after clicking on the Calculate
Distances button.
In contrast to the identity test, the Use Residuals option is not available during
cluster analysis. Spectral residuals are not taken into account when calculating
spectral distances. The calculation of factor spectra is not necessary during
cluster analysis, as test spectra will not be analyzed. To determine the spectrumto-spectrum distance (see section 6.1.2) only Z (covariance matrix) and L
(Eigen vectors) will be calculated. Click on the OK button to return to the
Parameters dialog box.
Figure 91: Cluster Analysis – Factor Spectra
7.10.4
Calculate Distances
Click on the Calculate Distances button to start the calculation of the spectrumto-spectrum distance. If the calculation has been finished, it is recommended to
store the method before you generate a dendrogram.
108
OPUS/IDENT
Bruker Optik GmbH
Cluster Analysis – Report
7.11
Cluster Analysis – Report
Click on the Report tab to have the analysis results displayed.
Figure 92: Cluster Analysis – Report tab
Use the drop-down list to define the form of report. You can have the results
displayed in the form of a dendrogram, histogram or diagnosis. Dendrogram is
set by default.
• Dendrogram
A dendrogram includes the spectral distances of all reference spectra.
Right click on the dendrogram and a menu pops up displaying
different options.
• Histogram
This kind of report is not intended to be used to analyze clustering.
Instead, the spectrum-to-spectrum distances between reference
spectra are analyzed. Such distances can be represented in the form
of a symmetrical n × n matrix (n being the number of reference
spectra).
The mean value and standard deviation are calculated, and the
distance values are displayed in the form of a histogram and divided
Bruker Optik GmbH
OPUS/IDENT
109
Reference Section
into classes. The first class, e.g. ranges from 0 to 1, the second class
includes spectral distances from 1 to 2 etc. Each class is represented
by a bar in the histogram. This bar indicates the percentage frequency
of spectral distances compared to the total number of distances
considered.
In this context Class means something different than in case of
clustering, where cluster can also be referred to as Class or Group.
Besides graphical representation, the Histogram includes statistical
information.
Figure 93: Cluster Analysis – Histogram
• Diagnosis
This view produces a horizontal cross section of the dendrogram.
Specify the number of classes to create a list which includes the
members of each single class. The spectral distance of the last
clustering will be displayed for each cluster.
110
OPUS/IDENT
Bruker Optik GmbH
Cluster Analysis – Report
7.11.1
Score Plot
The cluster analysis report can also be read out as score plots in 3D-format
which is indicated by the
tab. The additional Score Plot button will
only be displayed if you have selected Factorization as analysis method and
Diagnosis as report before.
Figure 94: Cluster Analysis - Score Plot button
Select Factorization from the Method drop-down list on the Parameters tab and
click on the Start Calculation button. Define at least 3 factors from the Factor
Spectra dialog which will serve as a basis for the 3D factor view. Subsequently,
select Diagnosis from the drop-down list and define the number of classes you
want to see in the score plot by using the Options button. For details on the
Options dialog see chapter 7.11.2. Click on the Score Plot button.
Figure 95: Cluster analysis - 3D factor view
Bruker Optik GmbH
OPUS/IDENT
111
Reference Section
If you position the mouse on one specific spectrum, the file name and group
name will be displayed. To improve the factor view, you can rotate the box. If
you position the mouse on the edge of the box, the cursor changes into
. To
rotate the box press the left mouse button and move the mouse to the position
desired.
Right clicking somewhere on the 3D view pops up the Properties button. If you
click on this button, the View properties dialog is displayed which allows
further plot settings.
7.11.2
Options
Click on the Options button to open the Cluster Analysis - Options dialog. You
can define the algorithm used to calculate spectral distances between different
clusters. In addition, you specify the Number of Classes used in the diagnosis,
and the parameters required for the histogram.
Figure 96: Cluster Analysis – Options
112
OPUS/IDENT
Bruker Optik GmbH
Cluster Analysis – Report
You can select between 7 different algorithms to calculate spectral distances
between clusters:
•
•
•
•
•
•
•
Single Linkage
Complete Linkage
Average Linkage
Weighted Average Linkage
Median Algorithm
Centroid Algorithm
Ward’s Algorithm
For details, see section 4.1.2.
Use the Dendrogram drop-down list to define the kind of labeling.
Dendrograms are labeled vertically. There are 4 possibilities:
•
•
•
•
File Name of the reference spectra
Sample Name of the reference spectra
File Number (file sequence in the list of reference spectra)
No Name Markers (no labelling at all)
A text file is automatically created for each dendrogram. This file has the same
name as the cluster analysis method and the extension *.DEN. The file includes
the dendrogram and exact clustering levels.
Specify the number of classes you want to test. If you test, e.g., original spectra
used to generate average spectra in an identity test, you have to enter the
number of average spectra into the Number of Classes field.
Define which part of the matrix you want to include into the histogram:
• Whole Matrix
In this case all distances will be used. With the matrix being
symmetrical and diagonal elements being 0, only a triangular matrix
without diagonal elements is used. The matrix size is
(n ⋅ (n – 1)) ⁄ 2 .
• Only Pairs (for repro tests)
The data record is divided into pairs and the distances between the
first and second spectra (first pair), third and fourth spectra (second
pair) etc. are calculated. The number of distances being considered is
n/2. This value can be used to determine the reproduction level of
measurements which have been repeated twice.
• Only Triplets (for repro tests)
Same as above, but this time the data record is divided in triplets. The
number of distances used for statistics is ( n ⁄ 3 ) ⋅ 3 = 3.
This value can be used to determine the reproduction level of
measurements which have been repeated three times.
Bruker Optik GmbH
OPUS/IDENT
113
Reference Section
• Only Reference (i.e. the last column)
This option only considers distances between the last spectrum
indicated in the list and all other spectra. The number of distances is
n-1.
• A Given Triangle
The distances are calculated for those spectra between k and l (rows)
of the list. This results in a triangle within the matrix. You have to
enter the k and l parameters. For k = 1 and l = n the result is identical
to the Whole Matrix option result.
• A Given Oblong
The distances are determined for those spectra between the k1 and l1
position (rows) and between the k2 and l2 position (columns) of the
list. This corresponds to a rectangle within the matrix.
The value specified in the Width for Classes of Distances field determines the
number of classes. The default value corresponds to a division of 20 classes, i.e.
the range from 0 up to the maximum distance is divided into 10 equal areas.
You can change this value. The maximum number of classes is 20. If you enter
an invalid number, the value will automatically be corrected.
The Window button opens the dendrogram, diagnosis or histogram within a
Report window. You can have the report printed out using the print options
from the OPUS Print menu.
7.12
3D File/Filelist
You can also set up a cluster analysis using either a 3D file or a file list. When
working with a 3D file you first have to load such a file. OPUS automatically
opens the 3D display indicated by the
114
OPUS/IDENT
tab.
Bruker Optik GmbH
3D File/Filelist
Figure 97: 3D view
For further details on the 3D window settings in OPUS refer to the 3D manual.
Now, select the Cluster Analysis command from the Evaluate menu and click
on the 3d File/Filelist tab. The following dialog opens:
Bruker Optik GmbH
OPUS/IDENT
115
Reference Section
A
Figure 98: Cluster Analysis - 3d File/Filelist tab
Drag & drop the spectra data block of the 3D file into the entry field (A in
figure 98). You cannot load more than one 3D file for a cluster analysis. To
remove the spectra data block, select the spectra file in the entry field and click
on the
button.
Now, click on the Parameters tab, define the frequency regions and select an
identification method. Click on the Start Calculation button. Depending on the
number of spectra the calculation procedure can take quite some time. If the
calculation has been finished, click on the 3D File/Filelist tab again. Define the
number of classes and click on the Store Trace button which is now enabled.
Figure 99: 3d File/Filelist - Defining number of classes
Note: If you activate the Ignore Rest check box, the biggest cluster will get the
number 0 in the trace report.
116
OPUS/IDENT
Bruker Optik GmbH
3D File/Filelist
The TRC data block (
) is added to the file displayed in the OPUS browser
window. Right click onto this data block to have the corresponding report
displayed.
Figure 100: Cluster analysis - TRC data report
As the TRC data report in figure 100 exemplifies the Cluster List column
includes the allocation to the classes defined before. To have the traces scored
in a 3D plot open the Map+Vid+Spec window by the New Registered Window
command from the Window menu. Drag & drop the TRC data block into the
first sub-window.
Bruker Optik GmbH
OPUS/IDENT
117
Reference Section
Figure 101: 3D plot of TRC data block
To have the clusters displayed in the second sub-window right click onto the
window and select the respective cluster list from the Select trace drop-down
list on the Mapping tab. For all the other plot options refer to the 3D manual.
118
OPUS/IDENT
Bruker Optik GmbH
3D File/Filelist
Figure 102: 3D plot of TRC data block with clusters
7.12.1
File List
If you use a file list several spectra of a particular spectrum type are combined
into one common file list. First, create a file list by the Setup File List command
in the Edit menu and store it. Drag & drop the LIST data block (
) into the
entry field (A in figure 98).
Now, click on the Parameters tab, define the frequency regions and select an
identification method. Click on the Start Calculation button. You can also store
traces before you have defined the number of classes on the 3d File/Filelist tab.
To open the TRC data block right click on the file list name in the OPUS
browser window and select Show Parameters from the pop-up menu. The TRC
data block is now added to the file list in the browser window. Click onto the
data block to be able to see the trace results.
Bruker Optik GmbH
OPUS/IDENT
119
Reference Section
7.13
Cluster Analysis – Store Method
Figure 103: Cluster Analysis – Store Method tab
This dialog box allows to store the cluster analysis method created. Click on the
Store Method button to open the standard Save File dialog box. The method file
has the extension *.CLA.
120
OPUS/IDENT
Bruker Optik GmbH
Index
Numerics
3D Files 114
A
A Given Oblong 114
A Given Triangle 114
Abs. Threshold 68
Add Average Spectra 6
Add Region 12, 86
Add Spectra 6
Artificial Spectrum 29
Average Linkage 27, 29, 30, 113
Average Spectrum 51
C
Calculate Distances 33, 108
Can Be Confused With 69, 74
Centroid Algorithm 113
Centroid Technique 30
Class Test 70
Class Test NOT OK 71
Class Test NOT PERFORMED 71
Class Test OK 71
Classes 25
Assign 9, 83
Clear Selected Regions 87
Cluster Analysis 25, 105
Performing 31
Clusters 25
Complete Linkage 30, 113
Composing 88
Confidence Band 44
Confidence Level 67
Conformity Data Block 49
Conformity Index 39, 46
Maximal 46
Conformity Index Limit 43, 45
Conformity Test 39
Performing 48
Report 49
Setup 39
D
Data Preprocessing 10, 61, 107
Data Processing 84
Dendrogram 25, 36, 109
Derivative 85
Detailed Report 94
Diagnosis 110
E
Eigen Vectors 23
Eigenvalues 23, 54
Euclidian Distances 28
Expected Reference 21
F
Factor Spectrum 57
Factorization 23, 28, 53, 88
Original Spectra 88
Factorization Method 58
File List 114
Fixed Algorithm 67
Frequency Regions 11, 60, 85, 107
G
Group Statistics 91
H
Histogram 28, 109
Hit Quality 19, 29, 51, 61, 87
I
IDENT Analysis 51
IDENT Report 21
Identified As 68
Identity Test 19
Result Display 102
Identity Test Limit 13
Identity Test Method 20
Identity Test Report 21
Interactive Region Selection 12, 85
L
Labels 36
Load Method 105
M
Main Library 4
Maximum Distance 90
Maximum Hit 90
Mean Distance 67, 91
Median Algorithm 113
Median Technique 30
U
N
V
No Reference Defined 104
Normalization to Reprolevel 28, 59, 88
Not Identified 69, 74
Validation 14, 74, 94
Report 15
Vector Normalization 61, 84
O
W
Only Pairs 113
Only Reference 114
Only Triplets 113
Ward’s Technique 29, 30, 113
Weight 59, 87
Weighted Average Linkage 30, 113
Whole Matrix 113
P
Parameters 10, 32
Pearson’s Correlation Coefficient 60
R
Reference Spectra 6
Reference Spectrum 31, 51, 106
Regions 85
Report 34, 109
Reprolevel 87
Result Report 94
S
Scaling to First Range 28, 59, 88
Score Coefficients 57
Score Plot 111
Second Derivative 85
Selectivity 95
Selectivity Histogram 95
Selectivity Report 94
Setup Identity Test Method 31, 77
Single Linkage 29, 113
Spectral Distance 25, 51, 60, 62
Spectral Residuals 54
Standard Deviation 67, 90, 91
Standard Method 21, 28, 51, 54, 88
Store Method 17, 100
Sub-Library 4, 81
Setting 7
Summary Report 94
T
Threshold 21, 67, 74, 90
TRC Data Block 117
Uniquely Identified 68, 74
Download PDF