Ground-truth RNAs for single-molecule nanopore sequencing prediction

Nanopore direct RNA sequencing (DRS) enables direct detection of N6-methyladenosine (m6A) in single RNA molecules. Most DRS methods do not capture m6A modification status on individual RNA molecules, a limitation stems in part from the lack of ground-truth training data with precisely known m6A status. To address this challenge, we synthesized ground-truth RNAs (GTRs) covering all combinations of 5-mer sequences with precise m6A status on each RNA molecule. Applying these GTRs to nanopore DRS we developed accurate models (NanoGTR) for predicting m6A sites and stoichiometry at single-nucleotide resolution in bulk transcriptomes and individual RNA molecules. To reveal compartmentalized m6A-dependent regulation on gene expression, we determined m6A stoichiometry and crosstalk patterns across transcriptome fractions from a human cell culture composed of total, chromatin-associated, nucleoplasmic, cytoplasmic, and polysome-associated RNA. Our analysis revealed differential m6A stoichiometry and correlation patterns between compartmental RNA populations that are relevant to polyA-length, alternative splicing, and translation efficiency. Our study provides a universal approach for constructing predictive models for RNA modifications at single-molecule resolution.