RNA-seq analysis in seconds using GPUs

We present a GPU implementation of kallisto for RNA-seq transcript quantification. By redesigning the core algorithms: pseudoalignment, equivalence class intersection, and the EM algorithm; for massively parallel execution on GPUs, we achieve a 30-50x speedup over multithreaded CPU kallisto. On a benchmark of 100 Geuvadis samples from Human cell lines the GPU version processes paired-end reads at a rate of 3.6 million per second, completing a typical sample in seconds rather than minutes. For a large dataset of 295 million reads, runtime drops from 40 minutes to 50 seconds. Our implementation demonstrates that careful algorithmic redesign, rather than naive porting of software, is necessary to fully exploit the computing power of GPUs in sequence analysis.