Discovery of Novel c-Jun N-Terminal Kinase 1 Inhibitors from Natural Products: Integrating Artificial Intelligence with Structure-Based Virtual Screening and Biological Evaluation

Abstract

c-Jun N-terminal kinase 1 (JNK1) is currently considered a critical therapeutic target for type-2 diabetes. In recent years, there has been a great interest in naturopathic molecules, and the discovery of active ingredients from natural products for specific targets has received increasing attention. Based on the above background, this research aims to combine emerging Artificial Intelligence technologies with traditional Computer-Aided Drug Design methods to find natural products with JNK1 inhibitory activity. First, we constructed three machine learning models (Support Vector Machine, Random Forest, and Artificial Neural Network) and performed model fusion based on Voting and Stacking strategies. The integrated models with better performance (AUC of 0.906 and 0.908, respectively) were then employed for the virtual screening of 4112 natural products in the ZINC database. After further drug-likeness filtering, we calculated the binding free energy of 22 screened compounds using molecular docking and performed a consensus analysis of the two methodologies. Subsequently, we identified the three most promising candidates (Lariciresinol, Tricin, and 4′-Demethylepipodophyllotoxin) according to the obtained probability values and relevant reports, while their binding characteristics were preliminarily explored by molecular dynamics simulations. Finally, we performed in vitro biological validation of these three compounds, and the results showed that Tricin exhibited an acceptable inhibitory activity against JNK1 (IC50 = 17.68 μM). This natural product can be used as a template molecule for the design of novel JNK1 inhibitors.

Machine learning in TCM with natural products and molecules: current status and future perspectives

“…Fu [95] combined CNN with traditional algorithm models to analyze tonguecoating properties and found that the performance of the integrated model was improved. On the basis of the voting and stacking strategy, Yang [96] performed rule integration-model fusion on three machine learning models of SVM, RF, and neural network, and achieved good performance. Ge [97] proposed an ensemble algorithm that integrated the attention mechanism and LSTM, and showed that this ensemble algorithm can effectively select salient locations with higher accuracy and less computation.…”