A hybrid unsupervised approach for accurate short read clustering and barcoded sample demultiplexing in nanopore sequencing

Abstract

Short nucleic acid sequences are usually attached as DNA barcodes for multiple sample sequencing and single cell protocols, which enables Oxford Nanopore sequencing to sequence multiple barcoded DNA samples on a single flow cell. However, due to the high base-calling error, short reads in Nanopore sequencing are difficult to be accurately identified by traditional tools. Here, we propose a hybrid unsupervised approach for the accurate clustering of short reads and demultiplexing of barcoded samples in Nanopore sequencing. In our approach, both the nucleic base information translated from base-calling and the raw current signal directly outputted by the flow cell are utilized. A GPU-supported parallelization strategy is proposed to ensure the runtime of our hybrid clustering. Comprehensive experiments demonstrate that our approach outperforms all the traditional unsupervised tools in short read clustering, and achieves comparable accuracy in barcoded sample demultiplexing compared with the learning-based methods.