Exploiting dependencies of patterns in gene expression analysis using pairwise comparisons


In using pairwise comparisons to analyze gene expression data, researchers have often treated comparison outcomes independently. We now exploit additional dependencies of comparison outcomes to show that those with a certain property cannot be true patterns of genes' response to treatments. With this result, we leverage p-values obtained from comparison outcomes to predict true patterns of gene response to treatments. Functional validation of gene lists obtained from our method yielded more and better functional enrichment than those obtained from the conventional approach. Consequently, our method promises to be useful in designing cost-effective experiments with small sample sizes. © 2013 Springer-Verlag.

Publication Title

Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)