Recursive Multi-view Integration for Subtypes Identification of Cervical Cancer

Multi-omics data provides an opportunity to develop an integrative computational method to identify clinically important cancer subtypes by looking into different omic-views simultaneously. Computationally there are two major challenges. First, the approximation of relevant subspace from each omic-view that holds the subtype information. And second, judicious integration of the subspace without any information loss. In this respect, a novel multi-omics clustering algorithm is presented in this study that integrates miRNA and mRNA expression, DNA methylation, and reverse-phase protein assays to identify clinically important subtypes of Cervical cancer (CESC). The proposed algorithm focuses on the statistical diversity present in each omic-view that directly impacts the sample similarities. A comparative analysis with eight other algorithms in this domain is also performed to demonstrate the clustering efficiency of the proposed algorithm and biological significance of the obtained subtypes. The method can be applied on other multi-omics cancer data-sets where subtype identification is important.