Fig. 3: Prediction of metabolic gene essentiality for higher-order organisms.
From: Accurate prediction of gene deletion phenotypes with Flux Cone Learning
a Receiver operating characteristic (ROC) curves of FCL model for Saccharomyces cerevisiae using a random forest classifier and 5-fold cross-validation; models were trained on N = 897 class-stratified gene deletions with q = 124 samples/cone computed from the Yeast9 genome-scale model33. Solid black lines are FBA baseline predictions computed for all genes in the GEM. b ROC curves of FCL model for Chinese Hamster Ovary cells using a gradient boosting classifier and 5-fold cross-validation; models were trained on N = 1832 class-stratified gene deletions with q = 127 samples/cone computed from a well-adopted genome-scale model32. In both panels, insets show confusion matrices for FCL predictions computed on class-stratified test genes held out from training (20%) and averaged across all folds. AUROC metrics for FCL models were computed as an average across folds ± one standard deviation. Solid black lines are FBA baseline predictions computed for all genes in the GEM. Details on model training can be found in the ”Methods”; additional classification metrics can be found in Supplementary Table S5. Measured (ground truth) essentiality labels were taken from the literature33,34. Source data are provided as a Source data file.
