Approved research

Evaluation of statistical learning to compute polygenic risk scores.

Lay summary

Polygenic Risk Scores (PRS) combine information from multiple SNPs into a single score for predicting disease risk. Current techniques to compute PRS make strong modeling assumptions. In parallel, statistical learning has been very active and successful at developing model-agnostic approaches for various predictive purposes. Based on several complex traits, our project seeks to evaluate and compare model-agnostic approaches to current techniques that compute PRS. The criterion for comparison is predictive accuracy in particular for individuals coming from underrepresented populations in large-scale genomic data. The results expected from the current proposal will generate novel computational techniques to improve prognostic for complex health-related traits. The developed tool will help to improve the prevention of a wide range of serious and life-threatening illnesses. For this reason, we think that our project is perfectly in line with the main UK Biobank?s purpose. Large-scale genomic data can be used to compute risk factors for diseases. Risk factors can then be used in population-based risk screening and stratification programs to identify at-risk individuals. The proposed project seeks to evaluate if modern machine learning methods can improve the predictive accuracy that is derived from genomic data. We require the full UK Biobank cohort subjects with complete phenotype and genotype information.