Improving polygenic prediction from whole-genome sequencing data by leveraging predicted epigenomic features

Area of Science:

• Genetics and Genomics
• Computational Biology
• Personalized Medicine

Background:

• Polygenic risk scores (PRS) estimate complex disease susceptibility using genetic variants.
• Whole-genome sequencing (WGS) enables large-scale detection of rare and de novo variants.
• Gene regulatory mechanisms are crucial for disease, yet often unintegrated into PRS.

Purpose of the Study:

• To develop a novel framework, Epi-PRS, for improved polygenic risk score prediction.
• To address limitations of existing PRS methods in handling nonlinear effects, rare variants, and regulatory context.
• To leverage large language models (LLMs) for integrating epigenomic data into PRS.

Main Methods:

• Developed Epi-PRS, a framework utilizing LLMs to impute cell-type-specific epigenomic signals from genotypes.
• Used imputed epigenomic signals as intermediates to model genotype-phenotype relationships.
• Validated the framework through simulation studies and application to UK Biobank data.

Main Results:

• Epi-PRS demonstrated improved predictive accuracy in simulations by incorporating nonlinear relationships, rare variants, and regulatory information.
• Epi-PRS significantly outperformed existing PRS methods for breast cancer and type 2 diabetes risk prediction in UK Biobank data.
• The framework enhanced both the predictive power and interpretability of PRS.

Conclusions:

• Epi-PRS offers a promising approach for more precise and biologically informed disease risk prediction.
• Integrating WGS data, epigenomic context, and LLMs advances personalized medicine.
• This framework contributes to a better understanding of complex genetic architectures and disease etiology.