Univariate-guided sparse regression for Biobank-scale high-dimensional omics data.

UniLasso achieves sparse, interpretable polygenic risk scores with comparable accuracy to Lasso in UK Biobank data, selecting significantly fewer variants.

• Why it matters: Efficient and accurate risk prediction models are crucial for understanding genetic influences on health, but existing methods often lack scalability or interpretability in high-dimensional genomic data.
• What they did: The authors developed uniLasso, a two-stage penalized regression leveraging univariate coefficients, and extended it to incorporate external summary statistics, applying it to over one million variants in the UK Biobank.
• The result: UniLasso provides predictive performance similar to standard Lasso while producing sparser models, enabling more interpretable genetic risk scores and competitive results against methods like PRS-CS and lassosum2.