Heterogeneous OMICS Data Integration for Prioritization of SNP Pairs with Epigenomic Similarity

Abstract

Reliable and effective data fusion and integration methodologies are crucial for the analysis of large-scale omics data in the fields of biological and medical studies and related pharmaceutical and life science industries. Currently, effective data analyzing approaches that can comprehensively integrate the large scale of multi-tissue type, multiple genome-wide assay data are limited. Thus, the results of large scale high-throughput sequencing experiments haven’t been leveraged and substantial biological information and knowledge is yet to be discovered. To these challenges, this dissertation proposes a systematic framework using bootstrap tests to integrate omics datasets and to prioritize the pairwise relationships between SNPs. Specifically, the proposed integration framework is first present to integrate complex heterogeneous data structure with Multiple Factor Analysis and Mahalanobis similarity measurements as epigenomic similarity. The omics datasets are often considered with distribution assumption of Gaussian for convenience, while some of them have the pattern of zero-inflated negative binomial distribution. The proposed integration approach firstly targets at the Gaussian distributed sequencing data, then extends to accommodate datasets following other distributions, with a detail discussed case of integrating zero-inflated negative binomial distributed datasets. Different measurements of epigenomic similarities have been comprehensively compared using Mahalanobis based and Euclidean-based distance/similarities. A series of simulation studies have also been conducted to demonstrate that the proposed integration framework can successfully prioritize SNP pairs with pair-wise epistatic interactions with different distribution assumptions.