Skip to main content
. 2017 Mar 6;13(3):e1005413. doi: 10.1371/journal.pcbi.1005413

Fig 5. Ensembles outperform individual GENREs when predicting gene essentiality.

Fig 5

We generated an ensemble of 51 GENREs by gap filling against 25 randomly-selected positive growth conditions, 10 negative growth conditions, and 80% of the reactions from the Model SEED draft network. We predicted gene essentiality in CF sputum medium and compared the predictions to in vitro gene essentiality data. We found that the “consensus” threshold (blue squares) achieved a ~20% increase in precision over the best individual GENRE and a ~100% increase in precision over the worst individual GENRE. Similarly, the “any” threshold (red circles) achieved a ~40% increase in recall over the best individual GENRE and a ~170% increase over the worst. Note the threshold-dependent tradeoff between precision and recall.