JiuCuo: PacBio HiFi Read Correction Method using Preassembled Contigs based on Deep Image Processing
Abstract
In biomedical field, as a fundamental technology to obtain genome sequence towards disease treatment, the PacBio HiFi sequencing technology emerged these years has greatly benefited genome analysis especially genome assembly. Even if the HiFi reads contains only 1% base errors, correction is crucial before assembly. Most of the current assembly methods contain an effective and efficient error correction module, but there is still some space for improvement: all-versus-one read alignment could be applied for accuracy, base errors and SNPs could be distinguished, and adapters could also be removed. To achieve these improvements, we design a HiFi read correction method called JiuCuo, which aligns the reads to preassembled contigs to do SNP-aware error correction and adapter removal. JiuCuo converts the aligned reads into images, then applies improved Inception-v4 model to distinguish base errors from SNPs, and also applies YOLO-v8 model combining DBSCAN and k-mer-based matching to detect adapters. In our tests, JiuCuo is accurate in SNP-aware error correction and adapter removal, and the JiuCuo corrected reads could be further assembled into contigs of higher quality than the preassembled ones. The JiuCuo software can be downloaded freely from https://github.com/liuj001/jiucuo.