PASTIS

PASTIS is a highly-parallel distributed protein sequence aligner. It relies on distributed sparse matrix and graph operations to scale expensive many-against-many sequence alignments. PASTIS uses novel techniques such as Blocked 2D Sparse SUMMA to reduce high memory overhead, custom matrix operations for load balancing and can take advantage of accelerators for faster alignments. In 2022, PASTIS was among the 6 finalists for the Gordon Bell Prize, completing a many-against-many search of 400 million protein sequences using over 20,000 GPUs in less than 3.5 hours and sustaining a rate of 690 million alignments per second. Accelerator-enabled GPU ports can be accessed under CUDA and HIP branches in the repository. Refer to these works for more details.