Biological Effects Pattern Recognition Tool Using Multivariate Statistical Data Reduction

We extend multivariate data reduction techniques developed in Phase I to the creation of a system for toxicogenomic fingerprinting and class prediction. Class discovery techniques explored in Phase I can generate a library of biomarkers for a number of toxic agents and exposures levels. The focus in Phase II will be on the development of class prediction and interpretation techniques that, based on a short list of biomarkers, can: 1) determine exposure vs. non-exposure; 2) identify the toxic agent; 3) determine the exposure level. The classification framework will make joint use of biomarkers derived from genomic, proteomic, and metabonomic profiles. The introduction of powerful literature mining techniques adds another dimension, by reinforcing experimental design with an interpretation layer. The benefits of text mining are two fold. In the short term, it can provide instantaneous feedback by linking highly correlated gene expression measures to known patterns of interactions. In the long term, it has the potential to relate groups of biomarkers to actual networks and mechanisms of action for a more reliable understanding of toxicogenomic fingerprints.