It is increasingly appreciated that long non-coding RNAs (lncRNAs) associated with alternative splicing (AS) could be involved in aggressive hepatocellular carcinoma. Although many recent studies show the alteration of RNA alternative splicing by deregulated lncRNAs in cancer, the extent to which and how lncRNAs impact alternative splicing at the genome scale remains largely elusive. We analyzed RNA-seq data obtained from 369 hepatocellular carcinomas (HCCs) and 160 normal liver tissues, quantified 198,619 isoform transcripts, and identified a total of 1,375 significant AS events in liver cancer. In order to predict novel AS-associated lncRNAs, we performed an integration of co-expression, protein-protein interaction (PPI) and epigenetic interaction networks that links lncRNA modulators (such as splicing factors, transcript factors, and miRNAs) along with their targeted AS genes in HCC. We developed a random walk-based multi-graphic (RWMG) model algorithm that prioritizes functional lncRNAs with their associated AS targets to computationally model the heterogeneous networks in HCC. RWMG shows a good performance evaluated by the ROC curve based on cross-validation and bootstrapping strategies. As a conclusion, our robust network-based framework has derived 31 AS-related lncRNAs that not only validates known cancer-associated cases MALAT1 and HOXA11-AS, but also reveals new players such as DNM1P35 and DLX6-AS1with potential functional implications. Survival analysis further provides insights into the clinical significance of identified lncRNAs.
Keywords: alternative splicing; gene-regulatory network analysis; hepatocellular carcinoma; integrative network analysis; long non-coding RNAs (lncRNA); multi-graphic random walk; random walk.
Copyright © 2020 Wang, Wang, Bhat, Chen, Xu, Mo, Yi and Zhou.