Table 2. Summary of accuracy measures on the positive and mixed datasets.
| Datasets | Positive | Mixed | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Tools | #Detected | TP | S (%) | P (%) | F1 | AUC | #Detected | TP | S (%) | P (%) | F1 | AUC |
| CIRI | 10714 | 10686 | 92.29 | 99.74 | 0.96 | 0.92 | 10850 | 10668 | 92.13 | 98.32 | 0.95 | 0.92 |
| CF | 8186 | 8109 | 70.03 | 99.06 | 0.82 | 0.70 | 8239 | 8109 | 70.03 | 98.42 | 0.82 | 0.70 |
| DCC | 7506 | 7460 | 64.43 | 99.39 | 0.78 | 0.64 | 7510 | 7460 | 64.43 | 99.33 | 0.78 | 0.64 |
| FC | 9085 | 9035 | 78.03 | 99.45 | 0.87 | 0.78 | 9795 | 9035 | 78.03 | 92.24 | 0.85 | 0.66 |
| SG | 11381 | 10677 | 92.21 | 93.81 | 0.93 | 0.89 | 12126 | 10308 | 89.02 | 85.01 | 0.87 | 0.84 |
| CE | 9970 | 9936 | 85.81 | 99.66 | 0.92 | 0.86 | 9972 | 9936 | 85.81 | 99.64 | 0.92 | 0.86 |
| MS | 8208 | 8168 | 70.54 | 99.51 | 0.83 | 0.70 | 8206 | 8159 | 70.46 | 99.43 | 0.82 | 0.70 |
| UB | 8434 | 7985 | 68.96 | 94.68 | 0.80 | 0.67 | 8517 | 7500 | 64.77 | 88.06 | 0.75 | 0.58 |
| KNIFE | 11406 | 11360 | 98.11 | 99.60 | 0.99 | 0.98 | 11819 | 11300 | 97.59 | 95.61 | 0.97 | 0.92 |
| PF | 10496 | 10458 | 90.32 | 99.64 | 0.95 | 0.90 | 10524 | 10465 | 90.38 | 99.44 | 0.95 | 0.90 |
| NCLS | 7218 | 7214 | 62.30 | 99.94 | 0.77 | 0.62 | 7220 | 7216 | 62.32 | 99.94 | 0.77 | 0.62 |
TP: true positives; S: sensitivity; P: precision; F1: F1 score; AUC: area under precision/recall curve; CF: circRNA_finder; CE: CIRCexplorer; FC: find_circ; MS: MapSplice; SG: Segemehl; NCLS: NCLScan; PF: PTESFinder; UB: UROBORUS. Note: there were a total of 11579 true positives in these two datasets. After we removed the ‘uncertain circRNA candidates’ listed in CYC & TJC’s comment paper, 3110 candidates instead of 3150 (probably a typo in their paper) were obtained after we merged the data from their DataSet S1 and Dataset S2 files.