Fig. 3: A statistical framework for m6A detection using direct RNA sequencing.
a Schematic diagram of proposed strategy to detect m6A sites using direct RNA sequencing. RNAs generated in WT cells (METTL3 positive) will contain m6A modifications, while RNA from METTL3 KO cells (M3KO) will not. As modified RNA passes through the pore there will be a higher error rate during base-calling compared to unmodified RNA. When compared to the reference transcriptome, the aggregate fold change in the Match:Mismatch ratio will be lower in at nucleotides containing m6A. b Proposed strategy for masking neighboring candidates. Here, three sites within five nucleotides of an AC produce significant G-test scores. All candidates are collapsed to the single candidate within five nucleotides giving the highest G-test statistic. Collapsed/masked candidates are analyzed for their distance to nearest ‘AC’ dinucleotide. When nearest ‘AC’ dinucleotide is within the five-nucleotide window (dictated by nanopore size) the candidate is shifted to the closest ‘A’ within an ‘AC’ core, if possible. c For each significant candidate site with a one-fold or greater difference in the match:mismatch ratio, the distance to the nearest AC motif was calculated and plotted (gray). This was repeated after masking neighboring candidates (blue) and for the 53 genome-level sites identified across all four comparisons (gold). d Comparisons between WT and M3KO (KO) datasets yields 184 putative m6A modified bases (post-collapse) of which 53 are consistently detected across all four comparisons. e The consensus five-nucleotide motif for putative m6A modified bases in the Ad5 transcriptome is predominantly comprised of four common DRACH motifs.
