Gene classification and regulatory prediction based on transcriptional modeling

We present a methodology that aims to elucidate regulatory mechanisms by grouping together genes which share the same regulatory network. In our method, we use multi-state partition functions and thermodynamic models to derive six distinct correlation classes that correspond to various protein-protein and protein-DNA interactions. We then introduce a novel biclustering algorithm that clusters together genes whose expression profiles exhibit the derived correlations in various conditions. The functional enrichment and statistical significance of the resulting clusters is evaluated by precision-recall curves and calculated p-values. Moreover, we analyzed the upstream regions of all genes that comprise each cluster, in order to verify that the derived correlation classes capture the expression of genes with common regulation. We have been able to identify over hundred strongly conserved sequences, among which eight match well-known regulatory motifs. Finally, further analysis of the identified conserved sequences provides not only an explanation of the classification performance, but serves also as an indicator of the regulatory coherence of various groups