Precise and Scalable Adaptive Sampling for Nanopore Sequencing
Nanopore DNA sequencing has shown to be an excellent tool for the identification of pathogens in clinical metagenomics samples. However, most of the sequenced DNA in such samples originates from the human host, which makes the detection of pathogens difficult or even impossible. With adaptive sampling there exists a method that can reject uninteresting or overrepresented sequences during the sequencing run, which results in an enrichment of the underrepresented DNA without applying time-consuming and expensive laboratory protocols. We developed a new tool for nanopore adaptive sampling that combines fast CPU and GPU base calling with read classification based on Interleaved Bloom Filters and k-mer matching statistics. ReadBouncer shows a higher read classification sensitivity than other state-of-the-art classification tools for adaptive sampling while retaining a high specificity. Our tool also improves classification performance and memory usage compared to the other tools.