Software Tool for Complex Biomarker Discovery

We propose development of a feature discovery algorithm for identification of multivariate (or composition) biomarker signatures from high-dimensional omics data. The intended algorithm would achieve high degree of accuracy in validation on independent data sets, and would be capable of describing continuous host responses to continuous variation of stimuli or insults. The initial development of the proposed algorithm was funded under Phase I of the Army SBIR grant. Our approach is simple and robust, relying on the first two statistical moments (mean and standard deviation) and plain logical inference, and not requiring additional assumptions. Currently the prototype is capable of deriving compact biomarker signatures from data sets with gene expression variation values for over 10,000 genes under over 1600 conditions, and achieves 81% prediction accuracy in validation experiments. The objectives of the Phase II include (1) completion of development and testing of the feature recognition algorithm, (2) conversion of it into production-quality software product suitable for use by the US Government, and (3) an investigation of additional commercial opportunities for this technology in the private sector.