Polygenic Risk Score in complex diseases

Abstract

Motivation: Plenty genome-wide datasets are produced from complex diseases by traditional GWAS studies, but they are limited. A new approach has emerged in the last decade, the Polygenic Risk Scores (PRS), to combine several SNP into a single predictor to try to explain the complex genetic behind diseases like Asthma or Autism Spectrum Disorders. Results: Here we analyse genome-wide data from these two diseases a compute PRS with three different approaches, PLINK’s method, a machine learning approach (biglasso) and a targeted-based method using SFARI database. We find that this kind of analysis are quite complex like the diseases they try to predict, and PRS only explain a very low percentage of the variance of the disease. The validation analysis we performed show us that the parameters used to compute the PRS have to be optimize using bigger datasets. We also used a machine learning approach (XGBoost) to impute the data in certain analysis.