pls2 regression as a tool for selection of optimal analytical modality
TRANSCRIPT
![Page 1: PLS2 regression as a tool for selection of optimal analytical modality](https://reader036.vdocument.in/reader036/viewer/2022081623/613d75e2736caf36b75d9887/html5/thumbnails/1.jpg)
PLS2 regression as a tool for selection of
optimal analytical modality
Michael Madsen and Kim H. Esbensen
W www.acabs.dk E [email protected]
- a closer look at beer brewing process analysis with
alternative spectroscopic data types
1
![Page 2: PLS2 regression as a tool for selection of optimal analytical modality](https://reader036.vdocument.in/reader036/viewer/2022081623/613d75e2736caf36b75d9887/html5/thumbnails/2.jpg)
Aim of the study
- To compare the performance of three different matured spectroscopic
techniques for beer brewing process monitoring (complex biochemical
matrix)
- To investigate inter-relationships between spectral data types using the
PLS2 algorithm
- To quantify the amount of unique information contained in each spectral
data type based on bilateral PLS2 (e.g. NIR vs. Raman and vice versa)
- To propose a dedicated PLS2 approach for selection of optimal analytical
modality for any given chemical matrix (biochemistry, environmental
engineering, bio energy, chemical engineering)
2
![Page 3: PLS2 regression as a tool for selection of optimal analytical modality](https://reader036.vdocument.in/reader036/viewer/2022081623/613d75e2736caf36b75d9887/html5/thumbnails/3.jpg)
Bilateral PLS2 approach (in development)
3
- Clean-up spectral data for gross outliers and other unrepresentative data.
Apply spectral transformation if feasible (e.g. MSC on Raman spectra)
- Perform PLS2 on e.g. MIR (X) vs. Raman (Y). Apply all relevant validation
methods to document their individual behaviour
- Switch the two spectral data types, Raman (X) vs. MIR (Y), and redo the
PLS2
- Visualise the joint vs. unique information (Validation Variance) for a
selected model complexity (no. of PLS components)
![Page 4: PLS2 regression as a tool for selection of optimal analytical modality](https://reader036.vdocument.in/reader036/viewer/2022081623/613d75e2736caf36b75d9887/html5/thumbnails/4.jpg)
Campus-scale brewery at AAU EIT (sponsored by Carlsberg)
Reference: Thygesen and Toft, Control of brewery, B.Sc. thesis, AAU EIT, 2006, p. 11
Cooling
Technical summary
2 fermentors approx. 150 L each
Pre-treatment vessels
Active cooling
Active heating
Ultra-filtration unit
Heat exchanger
4
![Page 5: PLS2 regression as a tool for selection of optimal analytical modality](https://reader036.vdocument.in/reader036/viewer/2022081623/613d75e2736caf36b75d9887/html5/thumbnails/5.jpg)
5
Plant monitored and controlled through LabVIEW™
Future options / ideas
Augment program with chemometric tools:
- online score plots
- early warnings (stuck fermentations)
- real time prediction
- automatic process sampling
(Hopefully) leading to:
- comprehensive process documentation
- better tasting campus beer…
![Page 6: PLS2 regression as a tool for selection of optimal analytical modality](https://reader036.vdocument.in/reader036/viewer/2022081623/613d75e2736caf36b75d9887/html5/thumbnails/6.jpg)
6
Short introduction to the main metabolism
Reference: The Alcohol Textbook, 4th edition (ed. by KA Jacques et al.)
The production path
Crushing/grinding of barley malt
Wort (sugar solution)
Filtration
Inoculation
Fermentation
Lager-process
Tapping
![Page 7: PLS2 regression as a tool for selection of optimal analytical modality](https://reader036.vdocument.in/reader036/viewer/2022081623/613d75e2736caf36b75d9887/html5/thumbnails/7.jpg)
7
Typical batch fermentation behaviour
Reference: The Alcohol Textbook, 4th edition (ed. by KA Jacques et al.)
![Page 8: PLS2 regression as a tool for selection of optimal analytical modality](https://reader036.vdocument.in/reader036/viewer/2022081623/613d75e2736caf36b75d9887/html5/thumbnails/8.jpg)
8
Spectroscopic implementation:
- A few issues worth addressing
Sampling
Heterogeneity
signal/noise
ratio
instrumental
stability
costs
investment
operation
analytical
sensitivity
![Page 9: PLS2 regression as a tool for selection of optimal analytical modality](https://reader036.vdocument.in/reader036/viewer/2022081623/613d75e2736caf36b75d9887/html5/thumbnails/9.jpg)
Key to unlocking information in signals: vibrational Eigen modes
Based on Classical Mechanics and Molecular Quantum Theory the
presence of signals from specific molecules in various spectral data
types can be predicted
Infrared spectroscopy represents molecular dipole-dipole interactions
9
Near infrared spectroscopy represents molecular vibrations, stretching,
and rotations
Raman spectroscopy represents scatter of incident laser light
In other words: the chosen spectroscopic techniques represent three
different physical phenomena
![Page 10: PLS2 regression as a tool for selection of optimal analytical modality](https://reader036.vdocument.in/reader036/viewer/2022081623/613d75e2736caf36b75d9887/html5/thumbnails/10.jpg)
Features of the chosen spectroscopic techniques
NIR
- Signals have low intensity
- Wide spectral absorption bands
- Very much affected by the presence of water
10
MIR
- Signals have high intensity
- Generally narrow absorption bands
- Some influence from presence of water and carbon dioxide
Raman
- Signals have very low intensity
- Affected by fluorescence (use excitation laser frequency close to NIR)
- Selective
- Less sensitive to presence of water
- Heats optical sampling area (high power laser excitation technique)
![Page 11: PLS2 regression as a tool for selection of optimal analytical modality](https://reader036.vdocument.in/reader036/viewer/2022081623/613d75e2736caf36b75d9887/html5/thumbnails/11.jpg)
11
Fourier Transform InfraRed spectroscopy
Spectral range
4 000 – 650 cm-1
Resolution
8 cm-1
Co-adds
64
Background
Ambient air
Temperature
Ambient (22 ºC 1 ºC)
Mode
Attenuated Total Reflectance
![Page 12: PLS2 regression as a tool for selection of optimal analytical modality](https://reader036.vdocument.in/reader036/viewer/2022081623/613d75e2736caf36b75d9887/html5/thumbnails/12.jpg)
12
Fourier Transform Near InfraRed spectroscopy
Spectral range
9 000 – 5 500 cm-1
Resolution
16 cm-1
Co-adds
64
Background
Ambient air
Temperature
Ambient (30 ºC 1 ºC)
Mode
Transmission
![Page 13: PLS2 regression as a tool for selection of optimal analytical modality](https://reader036.vdocument.in/reader036/viewer/2022081623/613d75e2736caf36b75d9887/html5/thumbnails/13.jpg)
13
Raman laser spectroscopy
Spectral range
3 000 – 200 cm-1
Resolution
10 cm-1
Integration time
1 s
Laser output power
30 mW
Temperature
Ambient (22 ºC 1 ºC)
Excitation wavelength
785 nm
![Page 14: PLS2 regression as a tool for selection of optimal analytical modality](https://reader036.vdocument.in/reader036/viewer/2022081623/613d75e2736caf36b75d9887/html5/thumbnails/14.jpg)
14
Wet chemical reference analysis
Analysed parameters
1. Maltotriose (not metabolised by the yeast applied)
2. Maltose
3. Fructose
4. Glucose
5. Sum of the above
6. Ethanol
Method
High Pressure Liquid Chromatography with Refractive
Index detection, HPLC-RI
Sample preparation
Dilution with acidic eluent and filtration
Analysis time
Roughly one hour per sample (analysed in duplicate)
![Page 15: PLS2 regression as a tool for selection of optimal analytical modality](https://reader036.vdocument.in/reader036/viewer/2022081623/613d75e2736caf36b75d9887/html5/thumbnails/15.jpg)
15
Typical sugar utilisation during a fermentation
![Page 16: PLS2 regression as a tool for selection of optimal analytical modality](https://reader036.vdocument.in/reader036/viewer/2022081623/613d75e2736caf36b75d9887/html5/thumbnails/16.jpg)
16
Sneak preview- more in the accompanying paper
![Page 17: PLS2 regression as a tool for selection of optimal analytical modality](https://reader036.vdocument.in/reader036/viewer/2022081623/613d75e2736caf36b75d9887/html5/thumbnails/17.jpg)
17
NIR (X) vs. MIR (Y) MIR (X) vs. NIR (Y)
Cal var.
# PLS comp.Val var.
# PLS comp
Cal var.
# PLS comp.
# PLS comp
Val var.
![Page 18: PLS2 regression as a tool for selection of optimal analytical modality](https://reader036.vdocument.in/reader036/viewer/2022081623/613d75e2736caf36b75d9887/html5/thumbnails/18.jpg)
18
NIR (X) vs. MIR (Y) MIR (X) vs. NIR (Y)
XX1: 98 %
YY1: 72 %
XX2: 88 %
YY2: 27 %
![Page 19: PLS2 regression as a tool for selection of optimal analytical modality](https://reader036.vdocument.in/reader036/viewer/2022081623/613d75e2736caf36b75d9887/html5/thumbnails/19.jpg)
19
PLS2 bilateral inter-calibration (A B)
Route diagram:
1) Is there a satisfactory model?
2) Validation is the answer! (X-val is the right thing HERE)
3) If the validation is acceptable: cal-variance (X)? : XX1%
4) Conclusion: XX1 % explains YY1 %
5)Reverse modelling (BA) and repeat procedure
6) XX2 % explains YY2 %
![Page 20: PLS2 regression as a tool for selection of optimal analytical modality](https://reader036.vdocument.in/reader036/viewer/2022081623/613d75e2736caf36b75d9887/html5/thumbnails/20.jpg)
20
Inter-calibration insight
1: proportion of the common modelling variance (XX %)
2: proportion specific modelling variance
XX1: 98 %
XX2: 88 %
common modelling variance specific
![Page 21: PLS2 regression as a tool for selection of optimal analytical modality](https://reader036.vdocument.in/reader036/viewer/2022081623/613d75e2736caf36b75d9887/html5/thumbnails/21.jpg)
21
Thank you for your attention
Avoid spurious correlations
- practice SAFE validation