Co-Regulatory Functional Module Detection Based on Affinity Propagation and Neighborhood Inflation

At co-regulatory level, transcription factors (TFs) and microRNAs (miRNAs) co-regulate the gene expression, and perturbation co-regulation can lead to a malfunctioning system and diseases. The co-regulatory functional modules (CRMs) combined by these three components can serve as building blocks of co-regulatory network which could cooperatively participate in post-transcriptional level. Identification of CRMs can help comprehend regulatory mechanism of complex diseases and reveal pathogenesis. However, recent researches about detecting regulatory modules involving a single kind of regulators might not completely bring to light complex regulatory mechanism. So, in this paper, we propose a novel computational framework called APNICRM which detects miRNA-TF-mRNA co-regulatory functional modules by affinity propagation (AP) and neighborhood inflation (NI). APNICRM firstly utilizes regularized least squares to construct a co-regulatory network from multiple sources of data: miRNA/TF/mRNA expression profiles and priori regulatory relationships (miRNA-mRNA/TF-miRNA/TF-mRNA). Then, in order to produce meaningful seeds for neighborhood inflation, affinity propagation clustering and modified neighborhood similarity will be adopted to generate seed sets. To further infer CRMs, we develop new greedily expanding strategies that optimize internal clustering fitness function. In addition, for improving the speed of operation, multi-threaded approach is executed for paralleling. Results on the ovarian cancer (OV) dataset and breast cancer (BRCA) dataset have shown that our proposed method is able to produce compact and meaningful functional modules highly relevant to the biological conditions. Comparing with existing methods, CRMs of APNICRM include more significant biological function GO-BP terms and KEGG pathways with higher enrichment scores. In addition, topological characteristics analysis and case study verify the effectiveness.