Evaluation of Dorado v5.2.0 de novo basecalling models for the detection of tRNA modifications using RNA004 chemistry

Abstract

Direct RNA sequencing with Oxford Nanopore Technologies (ONT) captures nucleotide-specific current signals that reflect both sequence and chemical modifications, offering the potential to detect RNA modifications directly from native RNA molecules. To interpret such signals, ONT provides modification-aware basecalling models that estimate the probability of selected modifications at each nucleotide. In May 2025, ONT released updated modification-calling models (Dorado v5.2.0) for pseudouridine ({Psi}), inosine, m6A and m5C, alongside new models for 2'O-ribose-methylations, necessitating independent validation. Here, we benchmark Dorado v5.2.0 against v5.1.0 using ex cellulo tRNAs from Schizosaccharomyces pombe, leveraging their well-defined modification landscape. We generated modification probability profiles at single-nucleotide resolution and quantified model performance using curated sets of annotated and validated modification sites. Our results reveal that, despite notable improvements in {Psi} detection, most modification callers remain challenged by the dense and heterogeneous modification environments of tRNAs. This work provides the first comprehensive evaluation of Dorado v5.2.0 on native tRNAs and establishes a methodological framework for benchmarking future ONT modification models in complex RNA modification contexts.