Improving Clinical Risk Models Through Integration of Polygenic Risk Scores and Omics

Abstract

Genome-wide association studies (GWAS) have significantly contributed to the identification of genetic variants by leveraging thousands of loci associated with complex traits and diseases, leading to breakthroughs in human genetics research. However, interpretation of these results is often difficult as the GWAS-identified variants (single-nucleotide polymorphisms [SNPs]) often have small effect estimates on target traits, limiting their direct application to disease research. Methods such as polygenic risk score analysis have been presented to enhance disease risk prediction, however, the clinical utility of these methods are controversial. Similarly, the identification of genes associated with likely causal SNPs is unclear as identified SNPs explain only a small portion of heritability. Genetic regulations of GWAS findings are often characterized using statistical methods such as transcriptome-wide association studies (TWAS), but these methods tend to ignore functional mechanisms that play critical roles in gene-regulatory pathways, resulting in loss of predictive power.

In this essay we present three integrative methodologies for improving disease risk prediction as well as identifying likely causal genes through incorporation of omics level data and GWAS results for complex traits. First, we propose an integrated PRS method that leverages multiple PRS methods, as well as multiple GWAS datasets, to improve coronary artery disease (CAD) risk prediction over clinically used prediction models (PCE). Our results indicate that our proposed method improves model discrimination for incident CAD cases over the PCE model while also improving risk reclassification in European populations. Second, we propose a Cauchy integrative model that aggregates gene expression levels, DNA methylation, and splicing event information to increase identification of likely causal genes in 24 GWAS complex trait datasets. Additionally, we extend our method to identify causal genes for lung adenocarcinoma risk. Finally, we implement our Cauchy integrative model to identify likely causal omics biomarkers for CAD and propose a penalized regression framework integrative PRS model to improve risk prediction in an African population when limited to a European GWAS and reference panel.