Computational learning of small RNA regulation in pancreatic cancer progression.

Based on the fast-accumulating genomics data, gene regulation network modeling has been the primary computational means to infer interactions between genes and their regulators, such as such Transcription factors (TFs) and microRNAs, to study (post-)transcriptional regulation in biological systems. However, learning dynamic behaviors from data in complex diseases constitutes a challenge task due to the chaotic interplay between regulatory mechanisms and the fact that those interactions vary over time. It is therefore crucial to build scalable learning models that can consider different types of regulations by leveraging heterogeneous data that captures each behavior while meaningfully treating each specie in its uniqueness. This work explores integrative approaches to gene network learning in human cancer, with a particular focus on microRNA-mediated regulation. Specifically, we introduce a learning framework that integrates expression and interactome information of RNA and fuses distinct graphical models to transform the prediction of static interactions into the identification of semi-conditional ones. Through analyzing data on human pancreatic cancer, we have identified distinct gene regulatory networks associated with four progressive stages, from which a list of 15 microRNA-gene interactions are found conditional to stages. The subsequent functional analysis reveals significant microRNA-mediated dysregulations in major cancer hallmarks, particularly in adaptive immune response and lymphocyte proliferation, which shed light on the pathological processes and regulatory roles that microRNAs play in the process of pancreatic cancer progression. We believe this integrative model can be a robust and effective discovery tool to facilitate the discovery of key regulatory characteristics in other complex biological systems.