Can Big Data Analytics Recapitulate Biology? A Survey of Multi-omics Data Integration Approaches

This chapter presents several approaches for analyzing multi-omics data, including factorization methods, message passing algorithms, methods for multi-block data analysis and generalized canonical correlation analysis, network-based methods, Bayesian methods and classification and regression algorithms. It describes methods that integrate multiple high-dimensional omics data sets without using additional phenotypic information on the biological samples. High-throughput molecular and cellular analytical platforms are inundating researchers with high-dimensional multi-omics data. Multi-block partial square maximizes the covariance between multiple data sets and a response matrix of interest. The presence of multiple data sets observed on the same set of individuals leads to the natural use of methods for the analysis of multi-block data. The hypotheses generated through multi-omics data integration must be followed up with experimental studies in order to isolate true biological relationships from purely spurious results. Variable selection can significantly improve the predictive performance and interpretability of classification algorithms.