Retargeting Tensor Accelerators for Epistasis Detection

The substitution of nucleotides at specific positions in the genome a population, known as single-nucleotide polymorphisms (SNPs), has been correlated with number important diseases. Complex conditions such Alzheimer's disease or Crohn's are significantly linked to genetics when impact multiple SNPs is considered. often interact an epistatic manner, where joint effect may not be simply mapped linear additive combination individual effects. Genome-wide association studies considering epistasis computationally challenging, especially performing triplet searches required. Some contemporary computer architectures support fused XOR and population count highest throughput operations part tensor operations. This article presents new approach for efficiently repurposing this capability accelerate 2-way (pairs) 3-way (triplets) detection searches. Experimental evaluation targeting Turing GPU architecture resulted previously unattainable levels performance, proposal being able evaluate up 108.1 54.5 tera unique sets per second, scaled sample size, searches, respectively.