Modulator Resolves Transcript-, Allele-, and Condition-associated RNA Modification Differences from Direct RNA Sequencing
Nanopore direct-RNA sequencing (DRS) reports RNA modifications on single molecules, but there is no comprehensive algorithm for processing, aligning, and differentiating base modifications across reads that are often 5'-truncated. Here we present Modulator, which assigns every read, truncated or not, to a fragmentform, defined by a shared transcript 3'-end and the splice-junctions upstream of it. Of note, Modulator builds fragmentforms from genomic alignments without a reference transcriptome, estimates per-base modification stoichiometry within each fragmentform, and classifies each stoichiometry difference as transcript-, allele-, or condition-dependent. In the control HG002 transcriptome, Modulator built 23,746 fragmentforms, with 15,620 candidate novel isoforms that carried poly(A) signals as often as previously annotated transcripts. Also, assigning 5'-truncated reads to fragmentforms nearly doubled the number of testable and differentially modified sites. m6A sites within 100 nucleotides (nt) of a splice junction or within 1 kb of a poly(A) site were less modified in fragmentforms that use that junction or poly(A) site, indicating that local transcript context shapes where m6A is deposited. From the same reads, Modulator called SNPs with 93% precision and no phasing errors against the Genome in a Bottle benchmark. It identified 131 SNPs that create or remove a modifiable nucleotide and 1,358 allele-specific modification differences, most involving SNPs more than 50 nt from the modified site. In cells infected with Zika virus, which remodels gene expression relative to uninfected cells, Modulator identified 1,253 transcript-independent modification stoichiometry changes. Of the m6A changes at sites measured by GLORI, 96% agreed in direction with single-nucleotide-resolution GLORI maps of the same samples. Thus, Modulator accurately separates changes in RNA modification from transcript-associated and allele-associated differences and can improve the utility of nanopore DRS. Availability and Implementation: Modulator is free and open source: https://github.com/Theo-Nelson/Modulator. It is written in Python and runs on Linux OS.