Publication

LOGAN: High-Performance GPU-Based X-Drop Long-Read Alignment

Abstract digital pattern, cover of the publication on LOGAN, GPU-based sequence alignment

Pairwise alignment can account for more than 90% of the runtime of key bioinformatics applications, and it is especially costly for long reads, which can span from one thousand to one million bases. LOGAN is the first GPU optimization of X-drop, a popular algorithm that searches only for high-quality alignments and stops early otherwise.

Using six NVIDIA Tesla V100 GPUs, it is up to 30.7x faster than state-of-the-art CPU software, with equivalent accuracy. Integrated into the long-read alignment software BELLA, it reduces total runtime by up to 10.6x.

Authors: Alberto Zeni, Giulia Guidi, Marquita Ellis, Nan Ding, Marco Domenico Santambrogio, Steven A. Hofmeyr, Aydın Buluç, Leonid Oliker, Katherine Yelick

Published in: IEEE International Parallel and Distributed Processing Symposium (IPDPS), 2020