Hierarchical multiblock PLS and PC models for easier model interpretation and as an alternative to variable selection

Abstract

In multivariate PLS (partial least square projection to latent structures) and PC (principal component) models with many variables, plots and lists of b loadings, coefficients, VIPs, etc. become messy and results are difficult to interpret. There is then a strong temptation to reduce the variables to a smaller, more manageable number. This reduction of variables, however, often removes information, makes the interpretation misleading and seriously increases the risk of spurious models. A better alternative is often to divide the variables into conceptually meaningful blocks and then apply hierarchical multiblock PLS (or PC) models. This blocking leads to two model levels: the upper level where the relationships between blocks are modelled and the lower level showing the details of each block. On each level, ‘standard’ PLS or PC scores and loading plots are available for model interpretation. This allows an interpretation focused on pertinent blocks and their dominant variables. Such blocking is natural and straightforward in spectroscopy (multivariate calibration), quantitative molecular modelling (e.g. CoMFA) and process modelling. The principles of hierarchical multivariate PLS and PC modelling are reviewed, some problems with variable selection are discussed and the approach is illustrated for a data set with around 300 variables and 500 observations taken from a residue catalytic cracker (RCCU) at the Statoil Mongstad refinery in Norway. © 1996 by John Wiley & Sons, Ltd.

Quantitative Metabolomic Profiling of Serum, Plasma, and Urine by 1H NMR Spectroscopy Discriminates between Patients with Inflammatory Bowel Disease and Healthy Individuals

“…This method is ideally used for analyzing variable-rich data sets. 31 The idea of hierarchical modeling is to group the variables for the purpose of improved clarity and interpretability to reveal how the different blocks (concentration data from different biofluids) are related, which blocks provide overlapping or unique information, and which biofluid measurements are most useful from a predictive viewpoint. This blocking leads to two model levels: the upper level where the relationships between blocks are modeled and the lower level showing the details of each block.…”

Relationship between Composition and Toxicity of Motor Vehicle Emission Samples

“…Overall, the results suggested that PCA/PLS can be useful for identifying composition–response associations for complex exposures even when the number of exposure cases is small. An alternative to grouping and variable selection is hierarchical PLS [described by Wold et al (1996)], which was used (not shown) to confirm the conclusions of the PCA/PLS results presented in this article.…”

Environmental Metabolic Footprinting: A novel application to study the impact of a natural and a synthetic β-triketone herbicide in soil

“…To identify the most influential biomarkers in the PLS-DA model, Variable Importance in Projection (VIP) scores were calculated (Wold et al, 1983(Wold et al, , 1996(Wold et al, , 2001. The VIP score is a parameter frequently used for the selection of variables, that corresponds to the sum of square of PLS weights for each variable (Andersen and Bro 2010;Chong and Jun 2005;Rajalahti et al, 2009).…”