Fig. 3: Comparative analysis of 6mA tools’ performance on site-level.
From: Comprehensive comparison of the third-generation sequencing tools for bacterial 6mA profiling
a Precision-Recall Curve (PRC) shows the overall detection performance of different tools. AP (average precision) values are indicated for each method. b Precision and recall values plotted against the logarithmic number of adenine sites for different detection tools. c Receiver operating characteristic (ROC) curve depicting the performance evaluation of six tools. Area under curve (AUC) values are shown for each tool. d True positive rate (TPR) and false positive rate (FPR) plotted against the logarithmic number of adenine sites. e The curve of the F1 score changes with the number of adenine sites included. f Heat map with the number indicated shows the optimal F1 score, the ROC value, and the AP value. Color intensity scales with numeric values, darker indicating higher values. g The curve of the F1 score changes with the modification fraction provided by SMRT and Dorado, indicating the single molecule level (per-read) accuracy. The ground truth dataset for Psph WT comprised all 6mA sites within type 1 and type 2 recognition motifs. h PRC with AP values. i Precision recall values with log-transformed adenine site count. j ROC curves with AUC values. k TPR and FPR values with log-transformed adenine site count. l The curve of the F1 score changes with the number of positive predictions. m Heat map with the number indicated shows the best F1 score reached, the ROC values, and the AP values. Color intensity scales with numeric values, darker indicating higher values. n The curve of the F1 score changes with the modification fraction provided by SMRT and Dorado, indicating the per-read accuracy. The ground truth dataset for Psph ∆hsdMSR comprised all 6mA sites within type 2 recognition motifs. The small line plots in (e) and (l) provide a zoom-in view of the F1 score change as the number of adenines increases, focusing on the top 10,000 predictions. All outcomes from Tombo_denovo, Tombo_levelcom, and Tombo_modelcom are processed with a 5-mer shift as indicated in Methods. Source data are provided as a Source Data file. The curves of different colors represent the results of predictions using different tools, marked at the bottom. TPR true positive rate. FPR false positive rate.
