Fig. 6: RNA modification landscape and interplay of neighboring RNA modification sites across human cell lines.

a Summary of read counts, number of detected modification sites, and Jaccard similarity of modification sites across nine human cell lines. b Length distribution of modification clusters. c Number of modification clusters stratified by the number of modification sites per cluster. d Venn diagram showing the overlap of modification types within all identified clusters. e Schematic of the expectation-maximization (EM)-based model used to estimate the co-occurrence of mutually exclusive interactions between neighboring modification sites. Local outlier factor (LOF) is used to assess the probability of each read being modified at each site, followed by statistical testing to evaluate the significance of modification interactions. f Proportion of co-occurrence events between different modification type pairs. Arrow thickness reflects the relative frequency of source-target modification interactions. g Example of two co-occurring m6A sites in the DNAJB1 transcript detected in IM95 DRS data and validated by HeLa m6A-SAC-seq. h Heatmap of enrichment score for splicing regulators and RNA modification-associated proteins (writers, erasers, readers; WERs) at isoform-specific, exclusively modified clusters. i, j Enrichment plots for ELAVL1 (i) and FMR1 (j) at isoform-specific m6A-m5C modification clusters. P values were calculated using a two-sided permutation test. k Genome browser view of the RBIS gene illustrating the relationship between RNA modifications and alternative splicing. Colors indicate predicted m6A (green) and m5C (purple) sites within exon 4, and dashed lines demonstrate the upstream alternative spliced exon. Source data are provided as a Source Data file.