Inferring condition-specific miRNA-Gene modules from miRNA and mRNA profiling data

Recently, a class of small RNA molecules, microRNAs or miRNAs, has attracted interest from researchers for their unique role in post-transcriptional regulation. Due to their distinct cell-type/tissue-specific expression patterns, it is of high importance to identify condition-specific miRNA-gene modules for a complete depiction of gene regulatory networks. In this paper, we propose a novel method to integrate miRNA and mRNA data to identify condition-specific miRNA-gene modules. Specifically, a Consistency-based Masking Nonnegative Matrix Factorization (CMNMF) method is developed to incorporate existing biological constraints (like the repression of miRNAs on potential target genes) with simultaneous miRNA and mRNA profiling data for an improved performance in module identification. The experimental results on simulation data show that the condition-specific modeling framework improves the performance in predicting miRNA-gene relationships. More importantly, application of CMNMF to human colon cancer data revealed a biologically significant miRNA-gene module, which contains four up-regulated miRNAs (miR-182, miR-183, miR-221 and miR-222) and six down-regulated target genes annotated as cytotoxity mediated by nature killer cells. The proposed method can also be applied to various biological conditions, even with limited number of samples, to elucidate miRNA-involved gene networks.