Boosting polygenic risk scores via distributional regression to uncover potential gene-environment interactions

• Applicants: Dr. Carlo Maj, Ph.D.; Professor Dr. Andreas Mayr
• Subject Area: Epidemiology and Medical Biometry/Statistics ; Human Genetics
• Term: since 2024
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 534238115

This project aims to identify candidate variants for gene-environment interactions by implementing a boosting distributional regression approach for polygenic risk score modelling. Polygenic risk scores are widely used in statistical genetics for predicting traits characterized by a complex genetic architecture. However, the models applied to derive polygenic risk scores primarily focus on predicting the mean of the phenotype without explicitly considering the effect on the phenotypic variance, which may reflect gene-environment interactions. Our project aims to develop a distributional regression approach to derive sparse polygenic models for both the mean and variance of the phenotype to have a more comprehensive representation of genetic susceptibility for a trait. By including a variance component in the polygenic risk score framework, we aim to develop tools to identify individuals who could particularly benefit from environmental changes, such as lifestyle interventions. The analysis will be based on the UK Biobank cohort, a large population-based dataset with deep-phenotyping data including regular longitudinal updates. Through phenome-wide analysis, we will explore within-trait and cross-trait correlations, detect shared variants, and investigate gene-based colocalization effects. Longitudinal examination of phenotypic values will enable a comprehensive understanding of genetic influences and responses to environmental changes. Our proposed distributional regression approach will make it possible to report uncertainty for phenotype predictions, enhancing the accuracy of individual risk profile evaluations. We will make the algorithm and polygenic risk score models publicly accessible by developing a dedicated portal for querying and visualizing phenome-wide results, fostering collaboration and sharing knowledge.

• DFG Programme: Research Grants