Signal level RNA modifications detection in eukaryotic ncRNAs

Eukaryotic transcriptomes contain dozens of covalent RNA post-transcriptional modifications but remain largely uncharted. Current transcriptome-scale mapping methods are labour intensive and have pitfalls including cross-reactivity and lack of reproducibility. ONT direct-RNA sequencing signals respond to RNA modifications although the available tools for signal-level detection are still limited in sensitivity and scope. We generated dRNA-Seq datasets from human and mouse samples, including conditions where several known epitranscriptomic writers were knocked-down. In addition, we in-vitro synthesized control transcripts for several non-coding. To perform comparative analyses of our datasets we developed Nanocompore, a program downstream of Nanopolish that compares samples at the signal level and identifies significantly altered positions corresponding to putative RNA modification sites. We validated the prediction accuracy and sensitivity of Nanocompore using in silico generated datasets. For N6-methyladenosine (m6A) a large proportion of the RNA modification candidates were validated with orthogonal methods such as miCLIP-Seq and meRIP-Seq. Altogether, our work shows the versatility, robustness and scalability of dRNA-Seq for RNA modification mapping