A major breakthrough in improving the success rate of new drug development is the search for new technologies in drug research and development, such as the rapidly emerging DNA-encoded library (DEL) and artificial intelligence (AI) technologies. Among these, the most well-known case of AI-assisted drug discovery (AIDD) is the discovery of DDR1 inhibitors. However, in this case, AIDD was based on the structural derivation of a large number of active scaffolds reported by previous researchers, making it difficult to break through the scaffold constraints and falling into fast-following. In the current application stage, the lack of high-quality chemical structure-biological activity (affinity) big data is the main obstacle preventing AIDD from more effectively accelerating the process of new drug discovery.

On the other hand, as a high-throughput screening tool for discovering First-In-Class (FIC) lead compounds, the combination of DEL and AI has begun to receive attention in recent years. However, there are very few publicly reported cases in the literature, and the reported cases only use DEL technology for the screening of relatively classical kinase and receptor targets. DEL combined with AI has not been fully applied to the discovery of potential FIC drugs, especially for difficult drug targets and targets with minimal or no known active molecules.

Recently, the Guangdong Small Molecule Drug Innovation Center, in collaboration with its incubated company Shenzhen NewDEL Biotech Co., Ltd., for the first time applied DEL+AI to difficult drug targets and targets with minimal or no known active structures. Targeting a well-known target in the field of tumor immunotherapy (TIGIT, a protein-protein interaction target, with very few known active structures), they successfully discovered active compounds with FIC potential. Based on the data obtained from DEL screening and initial optimization, they constructed an excellent AI model. The research results were published under the title "Discovery of TIGIT inhibitors based on DEL and machine learning" in the internationally renowned SCI journal Frontiers in Chemistry (impact factor 5.545, https://doi.org/10.3389/fchem.2022.982539, accessible for the original paper).

The study utilized AI technology to analyze the large data generated by DEL screening against the TIGIT target, which contained very few positive samples. Through comparative analysis, including a) fingerprint molecular generation, b) oversampling, c) undersampling combinations, etc., the proportion of positive samples was amplified. Eventually, based on the minimal amount of active compounds, two AI models, LightGBM and MLP, were successfully established. These models exhibited satisfactory performance on the test set (~75% high hit rate). This research will significantly promote the progress of drug discovery targeting difficult-to-drug targets or targets with a lack of activity data. It also demonstrates the promising application prospects of DEL+AI in the field of difficult-to-drug targets and targets with limited data availability.

In the future, NewDEL Biotech will continue to drive technological advancements in this field, establishing broader connections between chemical structure and biological activity (affinity or function). This will gradually elevate the starting point of new drug development, thereby shortening the drug development cycle and increasing the success rate of new drug development. Ultimately, it aims to realize the company's vision of "empowering new drug development with distinctive technologies to create affordable and effective medicines for the people."