Work

Statistical Methods for Multiple Phenotypes and Gene-Set Association Analysis

Year: 2019
Type: dissertation
Abstract: As association studies continue to advance, more efficient statistical methods are required to fully utilize existing data and to provide insight into genetic architecture of complex traits. Identifying association for a set of phenotypes or with respect to a set of variants can be particularly useful for understanding how biological networks might be affecting the patho-physiology of outcomes. In this dissertation, I attempt to develop computationally efficient statistical methods that facilitate insights into the mechanism of complex traits and understanding of their underlying biology. In Chapter II, I propose a generalized framework for gene-based tests with multiple correlated phenotypes. In genetic association analysis, a joint test of multiple correlated phenotypes can increase power to identify sets of trait-associated variants within genes or regions of interest. Existing multi-phenotype tests for rare variants make specific assumptions about the patterns of association with underlying causal variants and the violation of these assumptions can reduce power to detect association. In this project we develop a general framework for testing pleiotropic effects of rare variants on multiple continuous phenotypes using multivariate kernel regression (Multi-SKAT). To increase power of detecting association across tests with different kernel matrices, we developed a fast and accurate approximation of the significance of the minimum observed p-value across tests. To account for related individuals, our framework uses random effects for the kinship matrix. Using simulated and exome-array data from the METSIM study, we show that Multi-SKAT can increase power over single-phenotype SKAT-O test and existing multiple phenotype tests, while maintaining type I error rate. In Chapter III, I extend Multi-SKAT to a meta-analysis strategy, namely Meta-MultiSKAT, to combine results from several studies. Our method involves extracting score statistics and phenotype adjusted variant relationship matrix from individual studies which are then combined using a kernel that models the heterogeneity of effects between the studies. The proposed method accommodates situations where one or more phenotypes have not been observed in a particular study and studies have different correlation patterns among the phenotypes. With minor modifications our method can be used to test the combined effects of common and rare variants in a region, as well as incorporate functional information on individual variants. Meta-analysis of 4 white blood cell subtype traits from the MGI and SardiNIA studies show that Meta-MultiSKAT can identify associations which were not identified by existing methods and were not significant in individual studies. In Chapter IV, I propose a subset based approach for gene-set (pathway) association analysis using variant level summary statistics. Existing gene-set association (GSA) methods can have low statistical power when only a small fraction of the genes is associated with the phenotype and interpreting results in terms of the underlying genetic mechanism can be challenging since they cannot identify possible active genes within the set. For this, we propose a maximum-type statistic that selects the subset of genes with maximal evidence of association as the driver-genes and evaluates its significance using efficient simulation techniques. Through the analysis of summary statistics from the UK Biobank data for 1201 phenotypes with 10679 gene-sets, we show that our method can be used to identify novel associations across a large number of phenotypes and gene-sets.
Language: English
Subfield: Genetics
Open Access status: green
OpenAlex ID: W3008790571
Cites: 139
Cited by:
Related to: 10