The integration of interactions in polygenic risk scores (PRSs)

Abstract

Context: A polygenic risk score (PRS) aggregates the effects of many genetic variants (usually SNPs) across the genome. Each variant contributes a small amount to the risk or likelihood of a trait and effects are typically derived from GWAS. However, most classic PRSs are based on additive models and do not explicitly account for gene-gene (GxG) or gene-environment (GxE) interactions, which can influence this risk. These interactions could improve PRS prediction performance, population transferability, and clinical value.

Objectives: The purpose of this scoping review was to map the current state of the art in integrating GxG and GxE interactions into PRSs. It focusses on the research issues addressed, the methodologies used, and the impact of these interactions on PRS transferability and clinical utility.

Methods: This work followed the PRISMA-ScR guidelines for scoping reviews. A systematic search was conducted in the Pubmed bibliographic database using keywords related to PRSs and GxG/GxE interactions. After selection by title, abstract and then full text according to predefined criteria, 15 articles were included in the review.

Results: The 15 papers chosen were grouped into two categories: methodological approaches (creation of new integration methods) and applied approaches. The analytical approaches used include both standard statistical models and machine learning and deep learning methods for capturing complex interactions. Standard PRS calculation tools (e.g., PLINK) and machine learning libraries (e.g., scikit-learn) were used for advanced modelling. Overall, the incorporation of GxG and GxE interactions led to moderate predictive increases in certain situations and enhanced PRS transferability between groups, but the advantages remain variable.

Conclusion: Integrating G×G and G×E interactions into PRSs reveals non-additive effects overlooked by standard models, potentially improving prediction performance and transferability. However, methodological heterogeneity, limited statistical power, and the variety of biological environments restrict these advantages. Future opportunities include developing harmonised analytical frameworks and establishing large multi-ethnic cohorts with rich environmental data. Additionally, undertaking targeted systematic reviews will fully validate and quantify the clinical efficacy of interaction-enhanced PRSs.