Hunting complex differential gene interaction patterns across molecular contexts

Heterogeneity in genetic networks across different signaling molecular contexts can suggest molecular regulatory mechanisms. Here we describe a comparative chi-square analysis (CPχ(2)) method, considerably more flexible and effective than other alternatives, to screen large gene expression data sets for conserved and differential interactions. CPχ(2) decomposes interactions across conditions to assess homogeneity and heterogeneity. Theoretically, we prove an asymptotic chi-square null distribution for the interaction heterogeneity statistic. Empirically, on synthetic yeast cell cycle data, CPχ(2) achieved much higher statistical power in detecting differential networks than alternative approaches. We applied CPχ(2) to Drosophila melanogaster wing gene expression arrays collected under normal conditions, and conditions with overexpressed E2F and Cabut, two transcription factor complexes that promote ectopic cell cycling. The resulting differential networks suggest a mechanism by which E2F and Cabut regulate distinct gene interactions, while still sharing a small core network. Thus, CPχ(2) is sensitive in detecting network rewiring, useful in comparing related biological systems.