gpu-coloc achieves 1,000-fold speed gain for genetic colocalisation at biobank scale

Colocalisation analysis is a widely used statistical approach for assessing whether two genetic association signals — for example, a disease GWAS hit and an expression quantitative trait locus (eQTL) — are likely to share the same causal variant. As biobanks grow and molecular QTL atlases expand, the number of signal pairs requiring testing has become computationally intractable with existing implementations.

Mihkel Jesse, Ago-Erik Riet, and Kaur Alasoo, publishing in PLOS Genetics on 17 June 2026, introduce gpu-coloc: a GPU-accelerated reimplementation of the established coloc algorithm. By combining efficient data storage with parallelisation on graphics processing units, the tool achieves approximately a 1,000-fold speed increase over the standard implementation, making comprehensive pairwise colocalisation across millions of signals feasible in routine analyses.

The advance is primarily methodological. Large-scale colocalisation is increasingly central to fine-mapping, causal gene prioritisation, and the interpretation of GWAS results in the context of functional genomic data. A tool that removes the computational bottleneck could meaningfully accelerate these workflows in both academic and translational research settings.

The paper is peer-reviewed and published in PLOS Genetics. Source code and documentation are expected to accompany the publication. Researchers working with large biobank datasets or multi-trait eQTL resources are the primary intended audience.