Biological Effects Pattern Recognition Tool Using Multivariate Statistical Data Reduction

The application of modern high throughput genomic, proteomic, and metabonomic technologies to the field of toxicology will provide significant breakthroughs and advances. By simultaneously looking at data on the gene, protein, and metaboliteconcentrations, a more complete picture of cellular behavior can be determined. Using statistical and mathematical algorithms applied to high throughput data, patterns consistent with low level toxic exposure can be determined. In the current proposal, weintend to apply multivariate linear and non-linear statistical and mathematical techniques to find complex patterns in the data that are consistent with validated samples of low-level toxic exposure. By using supervised machine learning techniques,training data with solid cross-validated measurements will be used to distinguish between different levels of toxic exposure. By seamlessly applying these techniques to all the different data types, small correlations and patterns can be best measured andidentified. Upon completion of this SBIR project, we expect to have developed a prototype software tool that integrates the multivariate supervised methods into a useable software package.