Messenger-RNA Modification Standards and Machine Learning Models Facilitate Absolute Site-Specific Pseudouridine Quantification

Chemical modifications to mRNA are dynamic and are important for fine-tuning gene expression, but they are challenging to quantify due to low copy number and limited tools for accurate detection. Direct RNA sequencing (DRS) using nanopores has been used to assess the presence of pseudouridine modifications by detecting mismatch errors in basecalling. However, this approach is not quantitative and strongly depends on the sequence context. In this work, we combine direct RNA sequencing of synthetic RNAs bearing site-specific modifications and supervised machine learning models to achieve analytical quantification in DRS data. Our models reveal that the most important signal parameters for accurate classification of pseudouridine are sequence dependent. We used our models to quantitatively profile pseudouridine occupancy in several human mRNA sites across publicly available DRS datasets from various cell lines, and found that while occupancy levels are highly conserved for some sites, others vary across cell types. Our study motivates the development of a pipeline of models for a ground-truth control set of RNA molecules with site-specific modifications that will allow quantitative profiling of pseudouridine modifications in transcriptomes.