Table 1. Benchmark metrics for toxify compared to clantox and Toxclassifier.
The top portion shows averages and percentages (in parenthesis) of true and false positives (TP & FP) as well as true and false negatives (TN & NF). Additionally, the proportions for accuracy, specificity, sensitivity, balanced accuracy, negative predictive value, positive predictive value, F-score, and Matthew’s correlation coefficient are listed. The bottom portion shows computational performance in terms of CPU time in seconds and memory usage in megabytes. Asterisks indicate metrics in which toxify outperformed.
| clantox | ToxClassifier | toxify | |
|---|---|---|---|
| TP* | 147.8 (54.2%) | 152.8 (55.8%) | 209.6 (76.5%) |
| TN | 223.4 (81.8%) | 270.2 (98.6%) | 263.0 (96.0%) |
| FP | 50.6 (18.5%) | 3.8 (1.4%) | 11.0 (4.0%) |
| FN* | 126.2 (46.2%) | 121.2 (44.2%) | 64.4 (23.5%) |
| ACC* | 0.68 | 0.77 | 0.86 |
| SPEC | 0.82 | 0.99 | 0.96 |
| SENS* | 0.54 | 0.56 | 0.76 |
| BACC* | 0.68 | 0.77 | 0.86 |
| NPV* | 0.64 | 0.69 | 0.80 |
| PPV | 0.74 | 0.98 | 0.95 |
| F1* | 0.63 | 0.71 | 0.85 |
| MCC* | 0.37 | 0.60 | 0.74 |
| CPU (s)* | NA | 100.18 | 4.05 |
| MEM (MB)* | NA | 6,824 | 293 |