Fig. 2: Deep learning modeling reveals introns deletion revives m6A deposition at splice site adjacent exonic regions of internal exons.
a The m6A density of transcripts around original last exon start were compared between full length (black line) and all introns deletion (pink line). For all introns deletion, the input for iM6A is the nucleotide sequence of mRNAs (only exon sequences with no introns). b, c Positional plot of ΔProbability for the RAC sites located in the first (b, 1000 genes) or last (c, 1000 genes) 500 nt region of internal exons. Red, blue, and gray dots were those sites that had increased (> 0.1), decreased (< −0.1), or not change probability (|ΔProbability| <= 0.1) respectively by introns deletion. d–i Positional plot of ΔProbability for the RAC sites located in the first or last 200 (d, e, 1000 genes), 400 (f, g, 1000 genes), 500 (h, i, 2000 genes) nucleotide region of internal exons (exon length: <= 200 nt for (d, e), > 200 nt & < 400 nt for (f, g), >= 400 nt for (h, i)). Red, blue, and gray dots were those sites that had increased (> 0.1), decreased (< −0.1), or not change probability (|ΔProbability| <= 0.1) respectively by introns deletion. j, k Positional plot for the frequency of top 50 m6A enhancers (j), m6A silencers (k) in mRNA sequences around the RAC sites. Increase sites (red line, ΔProbability > 0.1), and no change sites (gray line, |ΔProbability| <= 0.1). l, m Box plot of PhyloP score of latent m6A sites or no change sites in near (l, p < 2.2e−16) or away (m, p = 2.1e−5) from splice sites (n = 200,565 or 339,928 for (l), n = 8727 or 89,545 for (m). The box represents the 1st to 3rd quartile with the median marked by a horizontal line. The P-values were calculated by two-sided Wilcoxon test (Significance: ***p < 0.001.).
