#39 · Tools & infrastructure

Scalable algorithms reduce runtime and memory for biobank-scale genetic analyses

The dominant thread is algorithmic scalability. Several items replace existing procedures with reduced-rank, incremental, GPU-parallel, or triplet-based formulations so that analyses run at biobank or tree scale. GCTAg reduces REML memory and MLMA runtime for agricultural GWAS using reduced-rank Woodbury acceleration. GROMTools, a C++ engine with an R interface, reports about 100-fold faster CPU time for individual-level GReX imputation from PLINK2 genotypes. The GPU polygenic scorer processes 3,202 1000 Genomes samples against 5,385 PGS Catalog models in 6.5 h and is validated against pgsc_calc. STELAR-X achieves O(nk) memory with GPU parallelism across 100,000 taxa. ProcessNets addresses a related problem for ensembles of biological networks using an nc-tree structure and incremental property computation, though it operates on networks rather than genotypes. What is new is the explicit targeting of biobank and 100,000-taxon scale.