The OPLS methodology for analysis of multi‐block batch process data

Abstract

With increasing availability of different process analysers multiple data sources are commonly available and this will impose new challenges and enable new types of investigations. The ability to separate joint, complementary and redundant information in multiple block data will be of increasing importance. In this study data from a batch mini plant were collected and O2PLS was implemented to facilitate a combined analysis of spectroscopic and process data. This enables assessment of both the joint and complementary variations in the respective data sets. The different types of variation that were separated were then modelled together to evaluate their individual correlation to a time response. By combining data of different origin an uncomplicated summary of the variation was accomplished and a deeper understanding of process interactions was gained. The analysis of separated variation with a response variable proved useful for verifying the supposed correlation between the joint variation and time.

Spatial Chemometrics and Comprehensive Chemical Imaging based Molecular Histopathology Delineates Anatomical Heterogeneity at Single Pixel Resolution

“…Thereby, systematic variability in two data blocks is partitioned into a joint predictive variation and into variation that is unique to each block of data (Figure 2c). [25][26][27] O2PLS has been used in numerous studies including technical applications to model pre-processing effects on spectroscopic data 28 , in batch process development 29 but also in systems biology to model variation between sets of "omics" data. [30][31][32][33] Here, we used O2PLS in a first application to assess the pixel-to-pixel correspondence between the two registered IMS datasets taking advantage of their multivariate nature (Figure 2c).…”

Advantages of orthogonal inspection in chemometrics

“…Gabrielsson et al . applied an orthogonal methodology to multiblock batch process data, being able to understand the origin of the orthogonal information. A recently published orthogonal method by Löfstedt and Trygg allows one to find the joint variation and the unique orthogonal variation in multiple (>2) block modeling.…”

“…It may be the result of different types of physical, chemical, biological, and instrumental factors . Examples include drifts in the results due to changes in operating or environmental conditions, e.g., pressure, temperature, humidity, and instrument‐specific factors ; physical factors, such as particle size, homogeneity and light scattering, chemical and molecular properties ; biological sample variation due to different animals, diets and time of analysis ; instrumental factors, such as nonlinear instrument responses, unknown baseline effects, and other foreseen or unforeseen anomalies .…”

Multiblock variance partitioning: A new approach for comparing variation in multiple data blocks

“…O2PLS (orthogonal PLS) is a modification of the OPLS method that provides separate models for both joint and unique (non-correlated/orthogonal) variations between two blocks of data with more than one variable in Y [10]. However, this method suffers from the before-mentioned scaling issues as all X-blocks must be augmented prior to the prediction of Y if the unique part of more than one X-block is to be evaluated.…”

“…Our approach of using individual PLS models is selected to avoid the block-scaling issue when dealing with multiblock data or when augmenting several X-blocks. The scaling-ofthe-blocks issue has been discussed in many papers where the observation is made that very different results and interpretations are found depending on the scaling method of choice for block-scaling dependent methods [3][4][5][6][7][8][9][10][11]18]. …”