Finemap-MiXeR: A variational Bayesian approach for genetic finemapping

Area of Science:

• Genetics
• Statistical Genetics
• Bioinformatics

Background:

• Genome-wide association studies (GWAS) identify genomic loci associated with traits, but these loci contain many correlated variants.
• Fine-mapping techniques are crucial for pinpointing specific functional variants within these loci.
• Existing methods may face computational challenges or limitations with increasing numbers of causal variants.

Purpose of the Study:

• To introduce Finemap-MiXeR, a novel method for fine-mapping causal variants from GWAS summary statistics.
• To control for linkage disequilibrium among genetic variants.
• To provide a computationally efficient and flexible fine-mapping tool.

Main Methods:

• Utilized a variational Bayesian approach and Evidence Lower Bound (ELBO) optimization.
• Employed the Adaptive Moment Estimation (ADAM) algorithm for optimizing the ELBO gradient.
• Validated the method using synthetic data and UK Biobank height data.

Main Results:

• Finemap-MiXeR demonstrated comparable or superior accuracy to existing methods like FINEMAP and SuSiE RSS.
• The method's computational complexity is independent of the number of true causal variants.
• No matrix inversion is required, enhancing computational efficiency.

Conclusions:

• Finemap-MiXeR is an accurate and computationally efficient tool for fine-mapping causal variants in GWAS.
• The method's flexible mathematical framework supports extensions to cross-trait and cross-ancestry fine-mapping.
• Finemap-MiXeR advances the ability to interpret GWAS findings and identify functional genetic variants.