ReadCurrent: A VDCNN-based tool for fast and accurate nanopore selective sequencing

Nanopore selective sequencing enables targeted sequencing of DNA of interest using computational approaches instead of conventional expressions such as multiplex polymerase chain reaction or hybridization capture. The speed and accuracy of algorithms and models for selective sequencing are crucial. Compared to the sequence-alignment strategy, using deep learning (DL) models without alignment can be faster to computationally determine target DNA, which can theoretically greatly increase the efficacy of selective sequencing. However, the application of DL models is hindered by their comparatively low accuracy. We proposed a DL-based tool, named ReadCurrent, for nanopore selective sequencing. To train and evaluate ReadCurrent, we generated datasets through nanopore sequencing of DNA or cDNA samples from Saccharomyces cerevisiae S288C, a microbial community standard of eight species, ten synthetic SARS-CoV-2 RNA standards, and the human cell line 293T. We also constructed two mixture samples using human DNA (293T cell line) and microbial community standard samples (ZYMO RESEARCH) with ratios of 4:1 and 9:1, respectively, for ReadCurrent adaptive sequencing.