Interpretable machine learning prediction and multi-objective optimization of chiral lactic acid production from...

Accurate prediction and control of chiral lactic acid (LA) production from complex waste streams remain challenging due to nonlinear metabolic interactions and process variability. Unlike previous machine learning (ML) studies that focused solely on predicting total lactic acid (T-LA), this study developed an integrated ML and optimization framework that uses substrate characteristics as input variables to construct predictive models for the two chiral isomers, L-lactic acid (L-LA) and D-lactic acid (D-LA), while also enabling prediction and optimization of T-LA. Supervised models (XGBoost, Gaussian Process Regression, Support Vector Machine, and K-Nearest Neighbors) were combined with unsupervised learning (cluster analysis), global sensitivity analysis (SHAP), and multi-objective optimization using the NSGA-II algorithm. Among the tested models, XGBoost achieved the highest predictive accuracy, with R² values of 0.90 for L-LA and 0.79 for D-LA during testing, and 0.95-0.98 during independent validation. Sensitivity analysis identified volatile fatty acids (VFA) and soluble chemical oxygen demand (SCOD) as the dominant factors governing isomer production, revealing inhibitory effects at elevated VFA concentrations. Optimization results defined operating regions enabling complete suppression of individual isomers or maximizing L-LA (23 gCOD·L^-1) and D-LA (10 gCOD·L^-1). In particular, D-LA inhibition was primarily controlled by increasing VFA, whereas L-LA inhibition depended on complex interactions between VFA and substrate composition. It should be noted that the predictive performance of this model depends on the range of the training data, and prediction bias may occur when input variables fall outside the training set's input distribution. Overall, the framework proposed in this study provides a reference for the prediction and decision-making in selectively regulating LA isomers and optimizing the conversion of waste into bioproducts.