Integrating Multiple Resources To Identify Transcriptional Cooperativity With A Bayesian Mixture Approach

In the human genome, genes manifest dramatic diversities in terms of expression levels and expression patterns. However, all genes are controlled by limited transcription factors. Transcriptional cooperativity has been proved to be the main mechanism of complexity and precision in regulatory programs. Although many data types generated from numerous experimental technologies are utilized in an attempt to understand combinational transcriptional regulation, complementary computational approach that can integrate diverse data resources and assimilate them into biological models is still under development.In this study, we introduce a novel Bayesian mixture approach for integrative analysis of proteomic, transcriptomic and genomic data to identify transcriptional cooperativity. The model evaluation based on golden standard dataset of HepG2 demonstrates distinguishable power of features derived from disparate data sources and their essentiality to model performance. Meanwhile, our Bayesian model outperforms individual feature prediction and straight Logistic regression classifier. The model application which contextualizes transcriptional cooperativity within hepatocellular carcinoma progression reveals carcinoma associated transcriptional cooperativity alterations. Furthermore, derived transcriptional cooperativity networks are highly significant in capturing validated cooperativity as well as revealing novel ones, from which we observe that the development and progression of carcinogenesis share highly common regulations meanwhile hold distinct transcriptional cooperativity. Five hub transcription factors could be regarded as pivotal factors to understand transcriptional regulatory mechanism controlling formative carcinogenesis. Also, we take specific cooperativity E2F1-TFAP2C VS E2F1-TP53 and specific transcriptional complex FOS-EP300-JUNB VS FOS-ETS1-NFKB1, all extracted from the transcriptional cooperativity networks, as the instances to investigate the potential molecular mechanism underlying the cancerization from cirrhosis to HCC.Our Bayesian approach is able to identify transcriptional cooperativity by comprehensive analysis of various data resources from multiple cascades. We anticipate this Bayesian approach is a promising application in identifying tissue or disease specific transcriptional cooperativity and will further facilitate the interpretation of underlying mechanisms for various physiological conditions.