Transcriptional Target Prediction Using Qualitative Reasoning

Transcription target prediction from functional genomics data often involves incorporating a conjunction of complex prior biological knowledge to the analysis. Unfortunately, typical prior hypotheses are qualitative rather than quantitative in nature. But, many qualitative biological hypotheses can be decomposed into a set of logic statements on binary outcomes. Here, we present a new method to convert qualitative statements into a collection of binary statements that in turn generates a partial ordering of outcomes, which can be tested using a semi-parametric isotonic regression. This semi-parametric approach yields a flexible but principled way of testing biological hypotheses. We applied this method to a published Arabidopsis microarray dataset to identify organ specific transcriptional target genes, and tested predictions independently using the AtGenExpress dataset. Our new algorithm performed comparably to published approaches and allowed rapid analysis of complex, multiple gene selection criteria