A fork of REGENIE 4.1.2 that extends the toolkit with nonlinear phenotype modelling for large-scale genome-wide association studies.
REGENIE is the standard tool for whole-genome regression on biobank-scale data, but it assumes linear relationships between genotype and phenotype. Many complex traits don't fit that assumption — nonlinear effects go undetected.
A fork of REGENIE 4.1.2 with a nonlinear calculation path added to the step 2 regression. The modification preserves the original tool's memory-efficient block-ridge design while allowing non-linear kernel functions to be swapped in.
Large C++ codebases built for performance have very little slack — every added allocation shows up in benchmarks. Getting the nonlinear path to run at biobank scale meant understanding the existing memory layout deeply before touching anything.