Computationally efficient whole-genome regression for quantitative and binary traits.

Area of Science:

• Genetics
• Bioinformatics
• Computational Biology

Background:

• Genome-wide association studies (GWAS) are computationally intensive, especially with large cohorts and multiple phenotypes.
• Accounting for sample relatedness and population structure further increases computational burden.
• Existing methods struggle with memory and speed for multi-trait analyses.

Purpose of the Study:

• Introduce REGENIE, a novel machine-learning method for whole-genome regression.
• Improve computational efficiency and statistical accuracy in large-scale genetic analyses.
• Facilitate parallel analysis of multiple phenotypes and accommodate complex genetic architectures.

Main Methods:

• Developed REGENIE, a machine-learning approach for whole-genome regression.
• Implemented a memory-efficient strategy loading only local genotype segments.
• Introduced an approximate Firth logistic regression for unbalanced case-control data.
• Leveraged distributed computing frameworks for scalability.

Main Results:

• REGENIE demonstrates substantial speed improvements over existing methods in multi-trait analyses.
• The method achieves significant savings in compute time and memory usage.
• Validated using the UK Biobank dataset with over 400,000 individuals.
• Maintained statistical efficiency comparable to traditional methods.

Conclusions:

• REGENIE offers a computationally efficient and statistically robust solution for large-scale genetic analyses.
• The method is well-suited for analyzing multiple phenotypes simultaneously.
• REGENIE enables more accessible and scalable genome-wide association studies.