Latent variables modeling in chemistry: two case studies

Nowadays, increasingly complex data tables are generated during various chemistry activities, and they are harder and harder to summarize and visualize without appropriate tools. Multi-Variate data Analysis (MVA) methods can be applied to extract latent information from data. Projection methods such as Principal Components Analysis (PCA), Partial Least-Squares to latent structures (PLS) regression and Orthogonal Projections to Latent Structures - Discriminant Analysis (OPLS-DA) are perfectly suited to this task. Here we present two applications of such approaches. a) The relationship between molecular structure and blood brain barrier permeation data will be investigated by using a Quantitative Structure-Properties Relationship (QSPR) approach. b) A metabonomics case study will be discussed, where three genetically distinct strains of mice are characterized by means of Liquid Chromatography - Mass Spectroscopy (LC-MS) data on urine samples, with the objective to find potential biomarkers.

Powerful visualization tools will be used in both instances to extract information and understand statistical models.