Fig. 2: Empathi performance and predictions for previously unannotated proteins.
From: Empathi: embedding-based phage protein annotation tool by hierarchical assignment
a Precision and sensitivity curves as a function of confidence (prediction probability) for Empathi binary models trained on the general functional groups. PVP phage virion protein. b Robustness analysis: F1 scores, precision and sensitivity of models on holdout PHROGs using a confidence threshold of 50%. Bars represent the average performance over 10 iterations. As no transcriptional regulators were predicted in two iterations, the precision for these iterations could not be computed (zero division) and was thus not included in the average score. c Predicted protein counts for each functional group in the fraction of the original dataset (INPHARED) that did not align to proteins with known functions in the PHROG database.
