ModkitOpt: An optimised workflow for RNA modification stoichiometry estimation and site calling from nanopore sequencing

Abstract

Despite enabling single-molecule detection of RNA modifications, nanopore direct RNA sequencing lacks standardised approaches for modification site calling. This requires accurately quantifying per-site modification stoichiometry and selecting an appropriate stoichiometry cutoff to classify sites. Here, we show that modkit, the de facto standard tool for estimating modification stoichiometry, is highly sensitive to parameter selection, and its heuristic parameter choice consistently leads to markedly sub-optimal site calling, which is exacerbated in datasets where dorado prediction confidence is heterogeneous. We also demonstrate that the choice of stoichiometry cutoff significantly affects false positive and false negative rates for called sites, leading to divergent biological conclusions. To address both limitations we introduce ModkitOpt, a pipeline that identifies then applies the optimal modkit parameters and stoichiometry cutoff for any modification type given a set of validated sites, producing optimised site calls for any nanopore sequencing dataset. Across multiple modification types and biological contexts, ModkitOpt consistently recovers the precision and recall of called sites, establishing a robust framework for standardised RNA modification stoichiometry estimation and site calling from nanopore direct RNA sequencing. ModkitOpt is available at https://github.com/comprna/modkitopt.