Efficient cross-trait penalized regression increases prediction accuracy in large cohorts using secondary phenotypes

Area of Science:

• Genetics
• Statistical genetics
• Bioinformatics

Background:

• Polygenic risk prediction (PRP) is crucial for understanding genetic contributions to complex traits.
• Existing methods often analyze traits independently, missing shared genetic influences.
• Large-scale biobanks provide unprecedented opportunities for multi-trait genetic analysis.

Purpose of the Study:

• To introduce a novel multi-trait penalized regression (CTPR) method for enhanced polygenic risk prediction.
• To leverage shared genetic effects across multiple traits for improved prediction accuracy.
• To develop a computationally efficient algorithm for application to biobank-scale data.

Main Methods:

• Proposed a novel cross-trait penalty function integrated with Lasso and minimax concave penalty (MCP).
• Developed a parallel computing algorithm for efficient computation on large datasets.
• Applied the CTPR method to UK Biobank GWAS data (~1M SNPs, N=456,837).

Main Results:

• CTPR demonstrated superior predictive performance compared to Multi-Trait Analysis of GWAS (MTAG).
• Prediction accuracy for height improved substantially when utilizing BMI information via CTPR.
• MCP+CTPR achieved R²=42.5% and Lasso+CTPR achieved R²=42.8%, compared to MTAG's R²=35.8% for height prediction.

Conclusions:

• CTPR is a powerful and practical approach for multi-trait polygenic risk prediction in large cohorts.
• The method effectively extracts information from secondary traits to improve primary trait prediction.
• CTPR offers significant advancements over existing methods for large-scale genetic studies.