Figure 3: Statistical learning refines predictions of the metabolic model. A regression model was devised for each amino acid. The variables μ, x, and x′ represent growth rate, predicted consumption rate and revised consumption rate, respectively. The regression coefficients were fitted from the training dataset. (a) Here we show the prediction qualities by fold change for a validation dataset before refinement by regression. The mean and variance of fold changes are comparable to the entire dataset (Figure 2b). (b) The prediction values were transformed by the regression model. Prediction of nearly all amino acids approach unity (red line). The datapoints are colored by the variance in fold change after transformation. Amino acids with high variance in fold change (y ≥ 1, red) – alanine, glycine, arginine and histidine – failed to be reliably corrected by the regression model.