Table 1. Accuracy metrics for blinded RDP prediction test.
| Rank | TPR (%) | MCR (%) | OCR (%) | UCR (%) | Acc (%) |
|---|---|---|---|---|---|
| Phylum | 99.5 | 0.1 | 41.1 | 0.1 | 98.8 |
| Class | 99.1 | 0.4 | 90.2 | 0.4 | 83.5 |
| Order | 97.8 | 0.7 | 84.2 | 0.7 | 92.9 |
| Family | 96.9 | 0.7 | 71.7 | 0.7 | 92.8 |
| Genus | 91.8 | 2.6 | 41.9 | 2.6 | 84.8 |
Notes:
See main text for definition of metrics. The low accuracy at class rank is striking; this is because there are many sequence with novel classes in v16 compared to v9 of the training set, and many of these (90%) are over-classified, i.e., falsely predicted to have class names which are known in v9.