Remora: a better way to mods


Detection of methylation from raw nanopore signal is an incredible strength of nanopore technology, but it has long been a cumbersome task to extract this information at the highest accuracy. The new Remora [link: https://github.com/nanoporetech/remora] modified base detection framework, presented here, will transform the way information on modified bases is extracted from nanopore data. Previously, the highest accuracy methylation calls were derived from a canonical basecalling model enhanced with modified base detection (via Megalodon or Guppy). Remora models, instead, separate canonical basecalling from methylation calling, thus enabling the highest quality canonical and methylation calls from a single run, and with minimal computational overhead. Remora models show improved performance over Megalodon methylation benchmarks and are much simpler to use and train. Remora models enable users to extract standard methylation with minimal effort, as well as enabling more adventurous users to investigate additional modified bases with much simpler training data. The simpler training data has enabled the training and release of a high-quality 5mC and 5hmC model, which enables a new frontier in biological discovery. Remora models trained on the latest R10.4 Kit 12 chemistry achieve the best methylation detection from nanopore data.

Authors: Marcus Stoiber