London Calling 2023: Dynamic, adaptive sampling during nanopore sequencing using Bayesian experimental design

Nanopore sequencers can reject molecules after analysis of a small initial part. Until now, selection has been based on predetermined regions of interest, which inhibits re-focusing on molecules that may contribute most to experimental success. We present a new method to generate dynamically updated targets by quantifying remaining uncertainty by streaming data to decide whether the expected information of a newly observed molecule warrants fully sequencing it. We illustrate this by mitigating coverage bias in a microbial community, improving variant calling. Further, we expand our method for true de novo enrichment, that is, without prior information about sample composition. We achieve this by constructing  and incrementally updating assemblies in real time, which are then used to reject over-represented  sequences, thus mitigating abundance bias without requiring input genomes. Overall, these data-driven updates are applicable to many scenarios, such as enriching regions with increased divergence or low coverage, or unknown species in mixtures.