A Platform for Profitable Growth

1
Raw material verification with
AssureID problems and solutions
Dr. Yaroslav Sokovikov Yuri Shishkin SchelTec AG
2
Introduction

• NIR Spectroscopy is a useful measurement
  throughout various stages of the manufacturing
  process, however it is particularly useful for
  raw materials checking and verification. If the
  materials to be identified are spectroscopically
  dissimilar, it is often only necessary to use a
  simple distance measure such as a spectral
  difference. If the spectra are similar, it may be
  necessary to utilize more sophisticated
  techniques which take into consideration both the
  variability of the spectra of interest and the
  differences between the spectra.
• The PerkinElmer Spectrum AssureID materials
  checking system provides pharmaceutical QA
  functions with a rapid, unambiguous means of
  verifying the identity and quality of a
  production material, confirming suitability for
  use in the manufacturing process. For the first
  time QA can implement a measurement system that
  integrates easily into the manufacturing process
  and return business benefits shortly after
  delivery.

3
AssureID key benefits

• Simple, proven spectroscopic technique with
  minimal sample preparation
• Choice of MIR or NIR systems
• Powerful chemometric engine eliminates the need
  for a chemometrician
• Designed to work the way QA works
• Minimal training requirements and up-keep
• 21 CFR Part 11 technical compliance
• Powerful trending and plotting tools

4
AssureID for NIR what is?

• PerkinElmer FT-NIR Spectrometer Spectrum 100N
  or
• FT-MIR/NIR Spectrometer Spectrum 400
• Plug-and-play accessory
• NIRA (NIR reflectance accessory)
• Fiber optic accessory
• Tablet checker
• Transmittance module
• Special software packed AssureID
• Method Explorer Editor
• Analyzer
• Result Browser

5
Problems

• Analysis of raw materials has show some problems
  in classification
• Has been received very low a level on the
  parameter Classification Rate
• Probably the reason of it in use of different
  parties of the same raw material
• The given presentation is devoted to methods of
  the decision of this problem

6
Calibration conditions

• Model calibration conditions

• Approximate number of samples is 350
• Each sample from a batch has three replicates
  scanned from different places of a same can by a
  fiber probe accessory
• Each scan was performed through two layers of PE
• No sample direct hit was allowed
• Spectrometer scanning parameters
• Range 100004100 cm-1
• Resolution 16 cm-1
• Apodization strong
• Accessory type fiber probe

• Model pre-process settings

7
Raw spectral data

• All the raw spectral data

• Visual analysis determines deviations in spectrum
  collecting process
• There were decided to separate all the data into,
  at least, two groups
• Main criteria baseline drift
• Drifting and sloping baselines determined BATCH
  ONE
• stable baseline determined BATCH
  TWO

• BATCH ONE and TWO are the subsamples of the
  same material

8
Batch 1 feature

• Batch one

• the current subgroup mainly consists of the
  samples which were suffered from the different
  pressure applied during probe scan
• the weak probe pressure affects the baseline
  nature
• SIMCA compensation is expected from the model
• baseline suffer samples included into the model
  are considered to protect the user from further
  sampling issues

9
Batch 2 feature

• Batch two

• the current subgroup mainly consists of the
  samples which were registered in the right way
  another user probably
• similar probe pressure provides the baseline
  nature stability
• SIMCA compensation - in case of the other user -
  is expected from the model
• right baseline samples included into the model
  are considered to protect the user from further
  sampling issues

10
Pre-processing data

• Pre-processing data a certain option provided
  by the software

• based on further calculations there were
  distinguished, that baseline slope and 9
  points 1st derivate provide the best calibration
  results
• the best recognition and rejection parameters are
  aquired by slope and derivate pre-processing
• MSC was selected as the most appropriate baseline
  normalization (SNV ended up to be a bit worth as
  the sample nature was a powder)

11
Baseline slope

• processed data by baseline slope all
  spectrums are divided in two kinds of a same
  material

• samples subgroup spread is visually evident (blue
  vs green)

12
Two submaterials models

• Calibrated variance space for the two
  submaterials of the same substance baseline
  slope

• Green is eventually a first group
• Blue is eventually a second group

• spectrum overlap is visually evident

13
Two submaterials validation

• Validation results

• Summary baseline slope
• the substance was divided into two different
  submaterials
• the separation was taken into account by
  processed data examination
• the validation process was performed on 49
  independent samples
• the results by baseline slope correction are
• Class 86
• Rejection 96

14
Preprocessed data 9 points 1st derivate

• processed data 9 points 1st derivate all
  spectrums are divided in two subkinds of a same
  material

• spectrum overlap is visually evident

15
9 points 1st derivate model

• Calibrated variance space for the two different
  submaterials of the same samples 9 points
  1st derivate

• Green is eventually a first group
• Blue is eventually a second group

• spectrum overlap is visually evident

16
9 points 1st derivate validation

• Validation results

• Summary 9 points 1st derivate
• the substance was divided into two different
  submaterials
• the separation was taken into account by
  processed data examination
• the validation process was performed on 49
  independent samples
• the results by 9 points 1st derivate are
• Class 78
• Rejection 96

17
Validation results

• Validation results short summary

• 9 points 1st derivate

• baseline slope

• baseline slope vs 9 points 1st derivate
• baseline slope and 9 points 1sr derivate
  assures good calibration and validation results
• baseline slope provides more accurate
  prediction - up to 10 better
• sample spread is more evident under baseline
  slope pre-process
• BS is selected as the best pre-process
  parameter for the further model configuration

18
Second calculation samples transfer

• All the agglutinated or the similar nature
  samples were decided to move into the second
  group

• Green is eventually a first group
• Blue is eventually a second group

• spectrum overlap is visually evident

19
Second calculation transfer and extreme samples

• All the red circled samples were decided to be
  separated into several subgroups blue circled
  samples were revised as an extreme group

• spectrum overlap is visually evident

20
Second calculation extreme samples

• Separation is complete black sphere means to be
  a variance space of the extreme samples which
  evidently overlap all the other spaces

• spectrum overlap is visually evident

21
Second calculation extreme samples

• Validation and Comments

• SIMCA model validation results are relatively
  poor corresponding to the Classification rate
• any other samples pre-processes do not help the
  result

• Solution

• exclude all the samples from the extreme -
  black - group

22
Second calculation extreme samples

• Extreme samples removal is done - all the data is
  now spread in a quite efficient manner

• However

• overlapping is occurred in pink and yellow
  zones

• overlapping is occurred in green and blue
  zones

• spectrum overlap is visually evident

23
Second calculation extreme samples

• Validation and Comments

• SIMCA model validation results are much better
  comparing to the previous calibration
• any other samples pre-processes and remodeling
  do not help the result

• slight overlapping in pink and yellow zones is
  not crucial as the samples are of the same
  material
• separate green and blue zones by checking the
  extreme samples in both spaces

• Solution

24
Second calculation working on blue zone

• Processing data

• blue space is now being adjusted by local
  extremes exclusion

25
Second calculation working on blue and green
zones

• Processing data

• green space is now being adjusted by local
  extremes exclusion
• exclusion is suggested by the interactive module
  of the software
• each sample is excluded manually though

• blue space is now being adjusted by local
  extremes exclusion
• exclusion is suggested by the interactive module
  of the software
• each sample is excluded manually though

26
Second calculation working on

• Developing and Validating

• no any spectrum overlap is visually observed nor
  detected by the software

27
Results

• Discussion

• Interactive SIMCA by AssureID software
• visual analyses in the beginning of the
  calibration helps to manage the samples
• software options provide RAW and PROCCESSED data
  acquisition on demand
• Assure ID helps the user interactively adjust
  the model parameters on any step of the
  calibration or the whole calculation may be
  completely automated
• integrated math database provides a brief course
  of SIMCA

• Overall performance
• first sample subgroups spread increased
  classification and rejection parameter
• second sample subgroups spread decreased class
  and rejection but further data processing
  assures better results
• model validation approves that all the pressure
  differences applied to the samples during data
  collection were compensated

• User satisfaction
• fast and easy way of model creation
• interactive and user-friendly interface of the
  software doesnt require a user to be a high
  grade specialist in SIMCA calculation