Improving GWAS discovery and genomic prediction accuracy in biobank data

Area of Science:

• Genetics
• Bioinformatics
• Statistical genomics

Background:

• Deep-phenotyped biobanks are crucial for genetic research.
• Efficient and accurate analysis methods are essential for leveraging biobank data.

Purpose of the Study:

• To apply and evaluate a novel Bayesian grouped mixture of regressions model (GMRM) for genomic prediction and genome-wide association studies (GWAS).
• To compare GMRM performance against existing methods in large-scale biobank data.

Main Methods:

• Bayesian grouped mixture of regressions model (GMRM) applied to UK and Estonian Biobanks.
• Comparison with annotation prediction models (LDAK, LDPred-funct) and BayesR.
• Extension of GMRM for mixed-linear model association (MLMA) SNP marker estimates in GWAS.

Main Results:

• GMRM achieved the highest reported genomic prediction accuracy across 21 heritable traits.
• GMRM outperformed other methods by 14-18% in prediction accuracy.
• GMRM-based GWAS identified 62-65% more independent loci than BoltLMM and Regenie.

Conclusions:

• GMRM offers a powerful and versatile approach for genomic prediction and GWAS in large biobanks.
• Accounting for minor allele frequency (MAF) and linkage disequilibrium (LD) differences improves genetic analysis.
• The findings highlight the importance of sophisticated modeling for genetic discovery and prediction.