Figure 2.

Validation results. For each of the five functional levels available in the FOAM ontology, three metrics were computed: recall (or sensitivity), precision (known also as ‘positive predictive value’ or sometime ‘specificity’) and F1-score (the harmonic mean of both). In all cases, precision stays >92% at the KO level to reach 97% at level 1 (21 classes). Recall varies much more, from 69% at the KO level, to 98% at level 1. Levels 2, 3 and 4 gave similar performance results for both recall and precision; and their F1-score ‘mean’ within a range of 92–94%.