DeepConsensus: Gap-Aware Sequence Transformers for Sequence Correction

Gunjan Baid; Daniel E. Cook; Kishwar Shafin; Taedong Yun; Felipe Llinares-López; Quentin Berthet; Aaron M. Wenger; William J. Rowell; Maria Nattestad; Howard Hua Yang; Alexey Kolesnikov; Armin Töpfer; Waleed Ammar; Jean‐Philippe Vert; Ashish Vaswani; Cory Y. McLean; Pi-Chuan Chang; Andrew W. Carroll

doi:10.1101/2021.08.31.458403

BioinformaticsbioRxivAug 31, 2021

Preprint

DeepConsensus: Gap-Aware Sequence Transformers for Sequence Correction

Gunjan Baid iD, Daniel E. Cook iD, Kishwar Shafin iD, Taedong Yun iD, Felipe Llinares-López, Quentin Berthet iD, Aaron M. Wenger iD, William J. Rowell iD, Maria Nattestad iD, Howard Hua Yang iD, Alexey Kolesnikov iD, Armin Töpfer iD, Waleed Ammar iD, Jean‐Philippe Vert iD, Ashish Vaswani iD, Cory Y. McLean iD, Pi-Chuan Chang iD, Andrew W. Carroll iD

Abstract

Pacific BioScience (PacBio) circular consensus sequencing (CCS) generates long (10-25 kb), accurate "HiFi" reads by combining serial observations of a DNA molecule into a consensus sequence. The standard approach to consensus generation uses a hidden Markov model (pbccs). Here, we introduce DeepConsensus, which uses a unique alignment-based loss to train a gap-aware transformer-encoder (GATE) for sequence correction. Compared to pbccs, DeepConsensus reduces read errors in the same dataset by 42%. This increases the yield of PacBio HiFi reads at Q20 by 9%, at Q30 by 27%, and at Q40 by 90%. With two SMRT Cells of HG003, reads from DeepConsensus improve hifiasm assembly contiguity (NG50 4.9Mb to 17.2Mb), increase gene completeness (94% to 97%), reduce false gene duplication rate (1.1% to 0.5%), improve assembly base accuracy (Q43 to Q45), and also reduce variant calling errors by 24%.

PDF DOI