Dissecting the multi-phenotype effects for cardiometabolic traits in highly dimensional whole genome and omics data through usage of multivariate analytical methods
File(s)
Author(s)
Anasanti, Mila Desi
Type
Thesis
Abstract
Over a decade, single-phenotype genome-wide association studies (SP-GWAS) have been used to identify the association between variants and cardiometabolic traits. Initially, our team performed an SP-GWAS meta-analysis of fasting insulin (FI) and fasting glucose (FG) of European and trans-ethnic ancestries within the Meta-Analysis of Glucose and Insulin-related traits Consortium (MAGIC). However, after calculating the variance explained, I found FG only slightly increased from MAGIC's previous analysis from 1.5% to 4.3%.
We proposed multi-phenotype GWAS (MP-GWAS) to boost the statistical power and performed MP-GWAS of fatty acids in NFBC1966 (N=4955) and replication in NFBC1986 (N=2687) to investigate fatty-acid metabolisms. The meta-analysis conducted by our team detected 10 signals associated with FAs (P<5x10-8) at PCSK9, GCKR, LPXN, FADS1, GPR137, ZNF259, LIPC, PDXDC1, PBX4, and APOE. For subsequent analysis, I proposed a new direct conditional analysis method within MP-GWAS, which detected multiple distinct signals within these loci.
While MP-GWAS is a powerful method for locus discovery, it could increase missing phenotype data drastically. I further investigated the properties of seven imputation methods within the MP-GWAS framework via an extensive simulation study.
I found that random forest (RF) is the best under various scenarios. However, as there was no available RF software designed for high-dimensional data, I developed the fastest to date RF imputation software, imputeSCOPA.
I applied imputeSCOPA to the NFBC data and performed an MP-GWAS of 31 metabolites on the imputed data and complete-cases (CC). I found that the analysis using imputed data boosted the power of MP-GWAS’s by identifying two novel signals at rs61803025 within FCGR3B (PCC=5.68 x10-7 vs Pimp=5.49x10-9) and rs181847072 within ADAMTS3 (PCC= 5.67x10-7 vs Pimp= 9.27x10-11).
These results demonstrate the increased power from MP-GWAS as compared to the traditional SP-GWAS. This work further highlights the importance of addressing missing data correctly and introduces a fast RF-based software imputeSCOPA.
We proposed multi-phenotype GWAS (MP-GWAS) to boost the statistical power and performed MP-GWAS of fatty acids in NFBC1966 (N=4955) and replication in NFBC1986 (N=2687) to investigate fatty-acid metabolisms. The meta-analysis conducted by our team detected 10 signals associated with FAs (P<5x10-8) at PCSK9, GCKR, LPXN, FADS1, GPR137, ZNF259, LIPC, PDXDC1, PBX4, and APOE. For subsequent analysis, I proposed a new direct conditional analysis method within MP-GWAS, which detected multiple distinct signals within these loci.
While MP-GWAS is a powerful method for locus discovery, it could increase missing phenotype data drastically. I further investigated the properties of seven imputation methods within the MP-GWAS framework via an extensive simulation study.
I found that random forest (RF) is the best under various scenarios. However, as there was no available RF software designed for high-dimensional data, I developed the fastest to date RF imputation software, imputeSCOPA.
I applied imputeSCOPA to the NFBC data and performed an MP-GWAS of 31 metabolites on the imputed data and complete-cases (CC). I found that the analysis using imputed data boosted the power of MP-GWAS’s by identifying two novel signals at rs61803025 within FCGR3B (PCC=5.68 x10-7 vs Pimp=5.49x10-9) and rs181847072 within ADAMTS3 (PCC= 5.67x10-7 vs Pimp= 9.27x10-11).
These results demonstrate the increased power from MP-GWAS as compared to the traditional SP-GWAS. This work further highlights the importance of addressing missing data correctly and introduces a fast RF-based software imputeSCOPA.
Version
Open Access
Date Issued
2020-09
Date Awarded
2021-03
Copyright Statement
Creative Commons Attribution Licence
License URL
Advisor
Froguel, Philippe
Publisher Department
Medicine
Publisher Institution
Imperial College London
Qualification Level
Doctoral
Qualification Name
Doctor of Philosophy (PhD)