Fig. 2: Algorithm performance varies depending on RNA modification type, modification stoichiometry and sequencing coverage.

A Dotplots of raw differential RNA modification scores along the modified 5-mer obtained by each software when detecting different RNA modification types. In the x-axis, 0 denotes the modified position. Position within the 5-mer of the ‘altered feature’ varies depending on the RNA modification type, but also depending on the software used to identify the modification changes. Data from three technical replicates from four different coverage levels is included. Box, first to last quartiles; whiskers, 1.5× interquartile range; center line, median. Source data are provided as a Source Data file. B Raw RNA modification scores along the E. coli 16S rRNA transcript for each individual software tested, obtained when comparing E. coli wild type and rsmF knockout strains. Analyses have been performed using either 100 (upper panel) or 500 randomly selected reads (lower panel). The modified site that is reported to be lost in the rsmF knockout strain (16S:m⁵C1407) is indicated with an asterisk (*). See also Figs. S3–S6. C Barplots of median differential modification raw scores obtained by each software in the modified 5-mer for each modification type present in the benchmarking. Raw scores are obtained by comparing the wild type and knockout strains, for each given site. Two randomly chosen rRNA modified sites that are not expected to change across any of the datasets have been included in the analyses as controls (16S:730 and 25S:2549). Error bars indicate median ± s.d. of three replicates. Source data are provided as a Source Data file. D Barplots of median raw scores in each modified 5-mer analyzed. Samples were analyzed using 1000 reads per sample, and with a diverse range of stoichiometry levels (see Methods). Error bars indicate median ± s.d. of three replicates. Source data are provided as a Source Data file.