Integrating Machine Learning to Explore Mechanisms of Danggui Liuhuang Decoction in Protecting Pancreatic β-Cells of Alloxan-Induced Type 1 Diabetic Mice

Dangguiliuhuang decoction (DD) is a conventional Chinese herbal formula used in treating type 1 diabetes mellitus (T1DM), although its specific mechanisms are still unclear. This research study explores the efficacy of DD to treat T1DM and its potential mechanisms of action. Furthermore, this study investigates the efficiency of DD in protecting pancreatic β cells in alloxan-induced diabetic mice utilizing machine learning models such as Random Forest (RF) and XGBoost (XGB). The performance of these models is compared with the other models, i.e., logistic regression (LR), gradient-boosted (GB), voting ensemble (VE), and support vector classifier (SVC). Results show that machine learning models analyzed the interaction between herb components and identified High_Ku and Medium_Dang as key predictors of β-cell protection. Ensemble approaches achieved the highest prediction accuracies (F1-Score: 1), while RF/XGBoost models were moderate in performance (F1-Score: 0.53-0.65). Experiments results show that high dose DD (3900 mg/kg/day) significantly reduced hyperglycemia. Machine learning models showed greater utility and impacts on traditional Chinese medicine (TCM) research through data limitation, affecting models' generalizability. Therefore, future research on biological topic topics could be undertaken with large-sized datasets and advanced deep-learning models.