Ultra-fast and High-quality Mapping of Error-prone Long Reads

To accelerate the mapping process of vast amounts of long reads to references, a novel mapper mapquik has an over 30 times speedup compared with the de facto standard minimap2 but maintains a comparable mapping quality on the human genome. However, the mapper is only available to accurately map PacBio HiFi reads with sequencing errors lower than 1%. Since the 3rd generation long reads with modest error rates higher than 1% are still widely used, like reads from Nanopore DNA sequencing technology, a versatile long-read mapper should consider more long-read cases. This paper adopts a mapping-friendly sequence reduction idea to compress sequencing errors from different sequencing technologies to boost seed sensitivity of mapquik. For reads with relatively high error rates, we combine an error-sensitive order from a deep learning algorithm to replace the random universe minimizer order in mapquik. An improved ultra-fast long read mapper named mapquikPLUS is verified to handle most 3rd generation read mapping tasks efficiently and a pipeline of mapquikPLUS followed by minimap2 shows better performance against other approximate mappers for ANI evaluation.