Ensemble Machine Learning- and Deep Learning-Driven Identification and Validation of Sennidin B as a Novel Dipeptidyl Peptidase-4 Inhibitor

Dipeptidyl peptidase-4 (DPP-4) is a key therapeutic target for type 2 diabetes (T2D). Several synthetic anti-DPP-4 drugs are currently available for the treatment of T2D; however, the need for safe and effective therapies remains unmet due to the side effects associated with existing DPP-4 inhibitors. This study aimed to integrate structure-based and machine learning (ML)-based virtual high-throughput screening to identify natural DPP-4 inhibitors. Random forest, logistic regression, support vector machine (SVM), and multilayer perceptron (MLP) models were trained on DPP-4 IC50 datasets. Among these, the SVM and MLP models achieved high predictive performance, with areas under the curve of 0.928 and 0.923, respectively. Screening of a natural compound database identified 107 compounds for further analysis. Subsequent structure-based screening, using sitagliptin as a positive control, identified sennidin B and doxorubicin hydrochloride as promising candidates with strong binding affinity for DPP-4. Molecular dynamics simulations (200 ns) and MM-PBSA calculations confirmed stable interactions with DPP-4. Further, sennidin B and doxorubicin hydrochloride inhibited DPP-4 activity in a concentration-dependent manner, with estimated IC50 values of 39.39 and 19.78 μM, respectively. Sennidin B also reduced DPP-4 mRNA and protein expression levels in Caco-2 cells. Overall, sennidin B shows promise as a natural DPP-4 inhibitor and warrants further investigation as a potential antidiabetic agent.