AI drug development is a cutting-edge field that combines innovative drugs with AI. With the rising popularity of ChatGPT, AI drug development has once again garnered widespread attention. The application of AI technology in drug development has seen significant growth in recent years as AI drug companies have shifted from conceptual discussions to technology validation and commercialization. This transition has led to a continuous influx of new drug development pipelines entering clinical research.

New drug development is a complex and lengthy process, encompassing various stages such as drug discovery, preclinical research, clinical research, and market approval. Currently, AI is primarily involved in the stages of drug discovery and preclinical research, which include target discovery, hit compound discovery, lead compound optimization, synthesis route optimization, ADMET prediction, crystal form prediction, and formulation design. Each application scenario corresponds to different biochemical datasets. In each unique scenario, the process involves problem set analysis, data integration optimization, algorithm model construction, and learning training. Key steps in this process include data annotation and algorithm model construction. However, the quality, quantity, and annotation of biochemical data continue to pose challenges for the industry.

Due to the scarcity of experimental data and the complexity of biochemical data, the effectiveness of annotation is often limited in AI pharmaceutical companies. As a result, the applicability of AI varies significantly across different scenarios. The differences between AI pharmaceutical companies primarily lie in their ability to develop various application scenarios and the degree of coverage of these scenarios. Some AI pharmaceutical companies focus on specific application scenarios, while others have the capability to deliver candidate compounds from a given target to preclinical trials. Among these scenarios, we believe that the greatest market demand lies in the prediction of ADMET properties, which include the absorption, distribution, metabolism, excretion, and toxicity of drugs. ADMET properties are closely linked to the efficacy and safety of drugs within the body and currently rely on animal experiments for verification. Therefore, accurate prediction of a compound’s ADMET properties during the drug discovery stage can effectively reduce research and development costs.

Furthermore, AI pharmaceutical companies have been actively investing in automated laboratories in recent years. The drug discovery process remains labor-intensive, and automated robots and instrumentation can alleviate the repetitive tasks performed by researchers. Laboratory automation can involve any experiment and every step, including chemical reactions, isolating proteins, culture cells, and dissecting animals, with the ideal future being full process automation. Currently, the reality is the automation of chemical synthesis. Several AI pharmaceutical companies have successfully implemented continuous operations for opening reactions, processing reactions, and purifying products 24/7, while also offering precise weighing, control of reaction temperature and time, automatic LCMS monitoring, and anhydrous and oxygen-free conditions. However, there are still obstacles to overcome, and the universality of automated laboratories is not yet fully realized. The application of automated laboratories will require a certain period of time for promotion, but it is undeniably a promising development that will replace manual experiments in certain scenarios in the future.