Fig. 1
From: Scaling computational genomics to millions of individuals with GPUs
Performance of GPU implementations for QTL mapping and signature analysis. a Average runtime to compute 10,000 iterations of Bayesian NMF using SignatureAnalyzer (SA) in R (gold) and SignatureAnalyzer-GPU (SA-GPU; purple). b Correlation heat map of mutation signatures derived from the R and GPU implementations of SignatureAnalyzer using the same input mutation counts matrix. c t-distributed stochastic neighbor embedding (t-SNE) of 1 million embryonic mouse brain cells. Colors indicate clustering based on SA-GPU decomposition performed in ~ 15 min. d Comparison of runtimes for cis-QTL (FastQTL on CPU (gold) and tensorQTL on GPU (purple)) and trans-QTL (tensorQTL on CPU and GPU). e GPU runtime of tensorQTL for the indicated numbers of samples and phenotypes. f Empirical cis-eQTL p values from the V7 GTEx release replicated using tensorQTL. Error bars indicate standard deviation of the mean