Skip to content

Sum of Single Effects (SuSiE) Model

Summary

The Sum of Single Effects (SuSiE) model is a powerful Bayesian regression framework designed for genetic fine-mapping. It reformulates the variable selection problem by assuming that the total genetic association signal at a locus is the sum of multiple single-effect signals, each representing exactly one causal variant. This formulation enables computationally efficient and accurate calculation of Posterior Inclusion Probabilities (PIPs) and Credible Sets.

Overview

Genetic fine-mapping aims to identify causal genetic variants (typically single-nucleotide polymorphisms, or SNPs) underlying genome-wide association study (GWAS) signals. Standard multiple regression approaches are often hampered by high linkage disequilibrium (LD), which creates strong correlations among adjacent variants.

The SuSiE framework addresses this by modeling the phenotype vector $y$ as: $$y = X \beta + \epsilon$$ where the effect vector $\beta$ is represented as the sum of $L$ single-effect vectors: $$\beta = \sum_{l=1}^L b_l$$ Each $b_l$ is a vector with exactly one non-zero entry, pointing to a single causal variant.

Extensions and Implementations

SuSiE serves as the foundational mathematical framework for several multi-ancestry and multi-trait fine-mapping methods: - SuSiEx: Extends SuSiE for cross-population fine-mapping using summary statistics. - SuShiE: Extends SuSiE for multi-ancestry fine-mapping of molecular quantitative trait loci (cis-molQTLs) using individual-level or summary data. - MultiSuSiE: Extends SuSiE to allow causal effect sizes to vary across ancestries, leveraging All of Us whole-genome sequencing data. - mvSuSiE: Extends SuSiE to the multivariate setting for joint fine-mapping of multiple traits.

Citations

  • Wang, G., Sarkar, A., Carbonetto, P., & Stephens, M. (2020). A simple new approach to variable selection in regression, with application to genetic fine mapping. Journal of the Royal Statistical Society: Series B (Statistical Methodology), 82(5), 1273-1300. DOI: 10.1111/rssb.12390