| L1 error | Precision | Recall/Sensitivity | F1-Score |
---|
MiCoP | 0.09124 | 1.0 | 0.98155 | 0.99069 |
Kraken | 1.15834* | 0.85147 | 0.90959 | 0.87957 |
MetaPhlAn2 | 1.24357 | 0.84388 | 0.369 | 0.51348 |
- Kraken is considered by its authors to be a read classification tool, not abundance estimation tool, so we put an asterisk next to its results. However, we note that abundance estimation is a common application for Kraken in practice. Overall, MiCoP outperforms the other two methods across all metrics. Kraken and especially MetaPhlAn are limited by the poor representation of viruses in their standard databases. L1 error is the sum of the absolute values of the differences between the computed species abundances and the ground truth species abundances. MiCoP’s L1 error was more than an order of magnitude better than the other tools, and MiCoP had the best precision and recall. * Based on read classification proportions; Kraken does not claim to perform abundance estimation