Combining artificial intelligence biophysical methods to look for non-native protein-protein interactions on anti-cancer drug development.
- Publication Date: 2020-04-03
| Execution Methods | The use of small molecules to influence the polymerization of proteins is a promising approach for developing new therapeutics. Most compounds created by structure-based design to manipulate the protein-protein interactions (PPIs) depend on detailed knowledge of the native interacting interfaces. In contrast, the compounds with therapeutic potential that target non-native PPI interfaces were discovered by chance alone. For the first time, we use the "non-native PPI interfaces " to establish a set of potential new drug development strategies and confirm the feasibility of using non-native PPI interfaces to target drug development. The current project aims to apply this concept to other disease-related proteins (such as cancer), thereby creating a new type of drug development platform. We performed the procedures to address this concept: (1) To identify non-native PPI interface via artificial intelligence methods. (2) To find potential compounds targeting non-native PPI interface via computational methods (3) To explore the mechanism and compound activity through biophysical assays, x-ray crystallography, and cell experiments. |
| Performance Evaluation | 1. In the first part, we used the unique algorithms, including support vector machines (SVM), deep neural network (DNN), and convolution neural network (CNN), to analyze the collected data covering cancer-related proteins, coronavirus-related proteins, and metabolic disease-related proteins. The selected interface features include the type of amino acid pairing, the number of pairs, the average distance, etc., a total of 210 components. In the end, the accuracy of predicting non-native PPI for specific protein categories can reach 98% accuracy, and the accuracy of predicting non-native PPI for all protein categories can get 85%. 2. We also continued to use the non-native PPI interfaces as the targets for developing anti-coronavirus drugs. This section successfully identified a lead compound acting on the CoV N protein and causing its abnormal aggregation. The manuscript has been prepared with relevant research results and ready for submission to the journal. |
| Conclusion & Suggestion | Since the PPI plays a vital role in both carcinogenesis and viral infection, the strategy of manipulating PPI via small molecules has been widely used in anti-cancer and anti-viral drug development. However, there is no successful example has been reported to discover drugs via targeting non-native PPI. For the first time in the current project, we used the "non-native PPI interfaces" to establish a novel platform for drug development. By integrating AI techniques, computational methods, and activity verification systems, we believe that our platform will open a new drug development milestone. |
| Appendix |