Table 2. Growth prediction accuracy in knockout models with constrained reaction directionality.

Outcome	Observed	Refined	Reference
TP	1673 (89.37%)	1580 (90.60%)	1678 (89.64%)
FP	28 (1.50%)	25 (1.43%)	9 (0.48%)
TN	18 (0.96%)	14 (0.80%)	37 (1.98%)
FN	153 (8.17%)	125 (7.17%)	148 (7.91%)
Total	1872	1744	1872
Sens	0.916	0.927	0.919
Spec	0.391	0.359	0.804
Prec	0.984	0.984	0.995
F1	0.949	0.955	0.955

Reaction deleted only if no functional enzyme complex associated with the reaction remains after knockout. Reconstructed model performance in predicting [34] “raw” observations (Observed) and “refined” observations (Refined) of [34]. Performance of the model of [34] (Reference) in predicting “raw” observations is also shown. Rows: Correct and incorrect growth predictions (TP, FP), correct and incorrect no-growth predictions (TN, FN), total number of cases, sensitivity, specificity, precision and F1 score.