SuShiE

Summary

SuShiE (Sum of Shared Single Effects) is an open-source Python tool designed for multi-ancestry statistical fine-mapping of molecular quantitative trait loci (cis-molQTLs). Built using JAX for hardware acceleration, SuShiE models effect size correlations across different ancestries to improve fine-mapping precision and compute ancestry-specific expression weights. It can analyze both individual-level phenotypes and summary statistics, running efficiently on CPUs, GPUs, and TPUs.

Method and Capabilities

SuShiE extends the foundational Sum of Single Effects (SuSiE) model by introducing a multi-ancestry framework: - Shared Causal Variants: It assumes causal variants are shared across ancestries but allows their effect sizes to covary according to a learned covariance matrix. - Ancestry-Specific Weights: By estimating ancestry-specific effect sizes, SuShiE outputs predictive weights that improve the power of downstream TWAS and PWAS. - Hardware Acceleration: Implemented in Python with JAX, enabling Just-In-Time (JIT) compilation and highly scalable performance.

Independent Benchmark

A head-to-head benchmark of multi-ancestry fine-mapping methods on real (non-simulated) TOPMed MESA pQTL data found SuShiE produced broader credible sets than SuSiEx, yielding higher recall (up to 0.219 against UK Biobank replication) but lower precision (0.105–0.137) — see Multi-Ancestry Proteome-Wide Association Studies for the full comparison against SuSiEx and MultiSuSiE.

Citations

• Lu, Z., Wang, X., Carr, M., Kim, A., Gazal, S., Mohammadi, P., Wu, L., Pirruccello, J., Kachuri, L., Gusev, A., & Mancuso, N. (2025). Improved multiancestry fine-mapping identifies cis-regulatory variants underlying molecular traits and disease risk. Nature Genetics, 57(8), 1881–1889. Source paper: s41588-025-02262-7.pdf