A mechanism-informed ML framework predicts polymeric long-acting injectable release with XGBoost (test R² = 0.9774) and finds predictive importance nearly uncorrelated with intervention effects (r = 0.132 and −0.021)
The study builds a mechanism-informed machine learning framework that chains a pharmaceutical-knowledge-driven directed acyclic graph, causal structure discovery (PC, NOTEARS, DirectLiNGAM), XGBoost release prediction, explainable AI (SHAP), ATE/CATE intervention-effect estimation, and counterfactual formulation analysis; on a public dataset of 181 release profiles, 3,783 fractional release measurements, and 43 drug-polymer pairs, XGBoost reached test R² = 0.9774, RMSE = 0.0491, and MAE = 0.0339, and correlation and SHAP importance agreed strongly (r = 0.761) while both agreed poorly with intervention-effect estimates (r = 0.132 and −0.021), indicating that variables useful for prediction differ from those that can be manipulated.