CuGen: A GPU-accelerated framework for large-scale genomics

genomics
UK Biobank
High-dimensional regression
penalized regression
Lasso
We present CuGen, a GPU-accelerated framework for biobank-scale genomics. UltraLasso, a hierarchical application of univariate-guided sparse regression, selects a compact active set of fewer than 30,000 variants for robust leave-one-chromosome-out confounding control, and the .cugen file format supports memory-optimized streaming and random access on GPU hardware.
Authors

Tuomo Kiiskinen

Joshua Richland

William Wang

Sophia Lu

Balasubramanian Narasimhan

Trevor Hastie

Robert Tibshirani

Manuel A. Rivas

Published

July 17, 2026