Fig. 3: NanoConsensus outperforms individual softwares in the prediction of RNA modifications. | Nature Communications

Fig. 3: NanoConsensus outperforms individual softwares in the prediction of RNA modifications.

From: Native RNA nanopore sequencing reveals antibiotic-induced loss of rRNA modifications in the A- and P-sites

Fig. 3

A Schematic overview of the NanoConsensus pipeline. B Per-position NanoConsensus scores along the 16 s and 18 s rRNA transcripts, for the battery of RNA modification types benchmarked in this work (m66A, Y, m5C, Am, Um and m7G), where the WT strain has been compared to a different knockout strain that lacks a specific rRNA modification (see Supplementary Data S1 for list of knockout strains used for each modification type). The known modified site that is lost is highlighted with an asterisk (*). Only one replicate is shown for each modification type, results obtained for additional replicates are shown in Fig. S15. C Barplot of NanoConsensus scores of the 5-mer regions centered in each modified site analyzed. Samples included in the analysis consist of 100% modified samples (WT), with varying read coverage levels, and are the same that were used to generate Fig. 2C. Error bars indicate median ± s.d. of three replicates. Source data are provided as a Source Data file. D Barplot of median NanoConsensus scores of the 5-mer regions centered in each modified site analyzed. Samples were analyzed with a range of stoichiometry levels (and fixed coverage of 1000 reads), which are the same samples that were used to generate Fig. 2D. Error bars indicate median ± s.d. of three replicates. Source data are provided as a Source Data file. E Radar plots of positive predictive values (PPV) for each software, RNA modification type and stoichiometry using either raw scores (upper track) or Z-scores (lower track). PPVs for raw scores were computed using the following threshold: for EpiNano, all sites reported as modified; for Nanopolish, sites whose Δmedian current intensity /median(Δmedian current intensity)>=5; for Tombo, sites whose p-value is lower than 0.01 after Benjamini–Hochberg correction; for Nanocompore, sites whose p-value is lower than 0.01 and with a GMM log odds ratio>0.5. PPVs for Z-scores were computed with a Z-Score threshold > 5. Source data are provided as a Source Data file. F Radar plots of positive predictive values (PPV) for NanoConsensus. PPV were computed using a Z-Score threshold > 5 and NanoConsensus score threshold > 5*(median across transcript). Source data are provided as a Source Data file.

Back to article page