Fig. 3: Testing on biological mRNA.
From: Detecting m6A at single-molecular resolution via direct RNA sequencing and realistic training data
a Full-transcript m6A sites - IGV snapshot of the HSPA1A gene in HEK293 cells, green shades indicate the read-level modification probability, \(P\left({m}^{6}A\right)\), of a single nucleotide on each read. b From read to site level - \(P\left({m}^{6}A\right)\) distributions of single nucleotides at 3 locations along HSPA1A, showing balanced (left), highly (center), and lowly (right) modified sites. The site-level stoichiometry, \(S\), is defined as the fraction of nucleotides with \(P\left({m}^{6}A\right)\ge 0.5\) at that site (see Methods). c Whole-transcriptome m6A profile - Site-by-site comparison of \(S\) predicted by mAFiA in HEK293 WT versus METTL3-KO, across all chromosomes. Site-density represents the number of sites within each 5% bin. n = 15316 sites. The red dashes mark the bins where \({S}_{{WT}}={S}_{{KO}}\). d Comparison with GLORI - Scatter plot of site-level m6A stoichiometry predicted by mAFiA (\({S}_{{mAFiA}}\), y-axis) versus values published by GLORI (\({S}_{{GLORI}}\), x-axis), across all chromosomes of HEK293 WT. Numbers in brackets are correlations between (\({S}_{{mAFiA}}\), \({S}_{{GLORI}}\)). GLORI does not report values of \(S\) below 10%. n = 5925 sites. e Titration experiment − 2d density-plots of site-level stoichiometry comparison between \({S}_{{mAFiA}}\) (y-axis) and \({S}_{{GLORI}}\) (x-axis), in 5 mixtures of HEK293 WT and IVT (WT fraction \({f}_{{WT}}={{{{\mathrm{0.00,0.25,0.50,0.75,1.00}}}}}\)). Red dashes correspond to the expected distribution of site-level stoichiometry depending on WT fraction: \({f}_{{WT}}\times {S}_{{GLORI}}\). f Slope extracted from linear regression of \({S}_{{mAFiA}}\) against \({S}_{{GLORI}}\), \(m\left(\frac{{S}_{{mAFiA}}}{{S}_{{GLORI}}}\right)\) (y-axis), in (e), as a function of \({f}_{{WT}}\) (x-axis). The observed variable, \(m\left(\frac{{S}_{{mAFiA}}}{{S}_{{GLORI}}}\right)\), as measured by mAFiA through the system-wide distribution of individual site stoichiometries, largely agrees with the underlying control variable \({f}_{{WT}}\). n = {2515, 3903, 6192, 5801, 5925} sites. Data are presented as fitted values +/− standard error. g Application to non-mammalian species - IGV snapshot of m6A sites detected by mAFiA (bottom) in Arabidopsis thaliana, juxtaposed with miCLIP peaks9 (top). h Site-by-site comparison of mAFiA-predicted site-level stoichiometries in wild type col0 (\({S}_{{col}0}\), x-axis) and mutant vir1 (\({S}_{{vir}1}\), y-axis) strains of Arabidopsis thaliana. The mutant strain shows significant down-regulation of m6A levels in otherwise highly modified sites. N = 11881 sites. Red dashes mark the bins where \({S}_{{col}0}={S}_{{vir}1}\). Source data are provided as a Source Data file.
