Statistical analysis for large-scale sequencing dataset using pathway information

Abstract

hence our method considers the correlation of pathways and handles an entire dataset in a single model. In addition, PHARAOH-multi further extends the original model into multivariate analysis, while keeping the advantages of our previous approach. We extend PHARAOH to enable analysis of multiple traits using hierarchical components of genetic variants. In addition, PHARAOH-multi can identify associations between multiple phenotypes and multiple pathways, with a single model, in the presence of subsequent genes within pathways, as a hierarchy. Through simulation studies, PHARAOH was shown to have higher statistical power than the existing pathway-based methods. In addition, a detailed simulation study for PHARAOH-multi demonstrated advantages of multivariate analysis, compared to univariate analysis, and comparison studies showed the proposed approach to outperform existing multivariate pathway-based methods. Finally, we conducted an analysis of whole-exome sequencing data from a Korean population study to compare the performance between the proposed methods with the previous pathway-based methods, using validated pathway databases. As a result, PHARAOH successfully discovered 13 pathways for the liver enzymes, and PHARAOH-multi identified 8 pathways for multiple metabolic traits. Through a replication study using an independent, large-scale exome chip dataset, we replicated many pathways that were discovered by the proposed methods and showed their biological relationship to the target traits.

In the past two decades, rapid advances in DNA sequencing technology have enabled extensive investigations into human genetic architecture, especially for the identification of genetic variants associated with complex traits. In particular, genome-wide association studies (GWAS) have played a key role in identifying genetic associations between Single Nucleotide Variants (SNVs) and many complex biological pathologies. However, the genetic variants identified by many successful GWAS have explained only a modest part of heritability for most of phenotypes, and many hypotheses have been proposed to address so-called missing heritability issue, such as rare variant association, gene-gene interaction or multi-omics integration. Methods for rare variants analysis arose from extending individual variant-level approaches to those at the gene-level, and extending those at the gene level to multiple phenotypes. In this trend, as the number of publicly available biological resources is increasing, recent methods for analyzing rare variants utilize pathway knowledge as a priori information. In this respect, many statistical methods for pathway-based analyses using rare variants have been proposed to analyze pathways individually. However, neglecting correlations between multiple pathways can result in misleading solutions, and pathway-based analyses of large-scale genetic datasets require massive computational burden. Moreover, while a number of methods for pathway-based rare-variant analysis of multiple phenotypes have been proposed, no method considers a unified model that incorporate multiple pathways. In this thesis, we propose novel statistical methods to analyze large-scale genetic dataset using pathway information, Pathway-based approach using HierArchical components of collapsed RAre variants Of High-throughput sequencing data (PHARAOH) and PHARAOH-multi. PHARAOH extends generalized structural component analysis, and implements the method based on the framework of generalized linear models, to accommodate phenotype data arising from a variety of exponential family distributions. PHARAOH constructs a single hierarchical model that consists of collapsed gene-level summaries and pathways, and analyzes entire pathways simultaneously by imposing ridge-type penalties on both gene and pathway coefficient estimates