Fig. 1

Overview of MLAGO, the machine learning-aided global optimization. Based on the enzyme information of EC number, Compound ID, and Organism ID, the machine learning model, which is trained and tested with 17,151 enzyme reaction data, predicts the values for unknown K_ms in a kinetic model. The predicted K_m values are used as the reference values in the constrained global optimization. Finally, a global optimization algorithm estimates K_m values so that the kinetic model fits experimental data. p is a set of K_ms to be searched, and q is log₁₀-transformed p. q^ML is log₁₀-transformed machine learning-predicted K_ms. p^L and p^U are the lower and upper bound vectors, respectively. Abbreviations: ML (machine learning), RMSE (root mean squared error), and BOF (badness of fit), AE (allowable error). For RMSE, BOF, and AE, see the main text