CCA based multi-view feature selection for multi-omics data integration

Recent technological advances in high-throughput omics technologies and their applications in genomic medicine have opened up outstanding opportunities for individualized medicine. However, several challenges arise in the integrative analysis of such data including heterogeneity and high dimensionality of the omics data. In this study, we present a novel multi-view feature selection algorithm based on the well-known canonical correlation analysis (CCA) statistical method for jointly selecting discriminative features from multi-omics data sources (multi-views). Our results demonstrate that models for predicting kidney renal clear cell carcinoma (KIRC) survival using our proposed method for jointly selecting discriminative features from copy number alteration (CNA), gene expression RNA-Seq, and reverse-phase protein arrays (RPPA) views outperform models trained using single-view data as well as three integrated models developed using data fusion approaches including CCA-based feature fusion.