Sparse Linear Algebra Accelerates Genotype Representation Graph Computation at Biobank Scale
Biobank-scale genomic analyses are increasingly constrained by computational costs, as hundreds of thousands to millions of samples and variants must be analyzed together. The genotype representation graph (GRG) compactly encodes population genetic variation to accelerate computation, but the current approach does not exploit modern accelerator architectures. This work introduces Mikado, a new…
Biobank-scale genomic analyses face computational constraints due to the need to analyze hundreds of thousands to millions of samples and variants simultaneously. The genotype representation graph (GRG) is a compact way to encode population genetic variation, but it fails to leverage modern accelerator architectures. This research introduces Mikado, a method for expressing GRG-based computation using sparse linear algebra primitives.
Mikado employs a reverse topological ordering of the graph nodes, ensuring that the GRG adjacency matrix is strictly block-lower-triangular. This allows the genotype matrix-vector product to become a sparse triangular solve, which can be broken down into a pipelined sequence of blocked sparse matrix-vector multiplies. By separating computation from graph representation, Mikado uncovers fine-grained parallelism and enables the utilization of hardware-optimized sparse primitives on GPUs.
Compared to the original GRG traversal approach, Mikado achieves an order-of-magnitude speedup and cost savings for PCA and BOLT-LMM, even when applied to All of Us cohorts. The methodology provides a scalable, hardware-portable, and researcher-friendly tool for population genetics at biobank scale.
Written by urgent.news from bioRxiv's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
