Better complex-trait prediction through nonlinear spatiotemporal factors

Genetic models mostly use only limited linear covariate adjustment. A peer-reviewed Nature Communications study by Ross DeVito and Melissa Gymrek of the University of California San Diego, published August 29, 2026, adds location, time and nonlinear interactions through gradient-boosted null models.

Across 16 UK Biobank phenotypes, the authors report a median 7.3 per cent R-squared gain for one downstream method. A small workflow addition could reveal environmental effects.

The metric is author-reported in a specific biobank; external population validation is needed.