The US Food and Drug Administration (FDA) recently issued a notice that it will gradually eliminate animal testing requirements in drug applications for monoclonal antibodies and other drugs, and instead use "emerging alternative methods" such as artificial intelligence (AI) prediction models, organoids, and organ chips that are more efficient and closer to human reactions. Is China's pharmaceutical industry ready to bid farewell to the historical stage of animal experimentation? What is the progress and reserve of related scientific research? At the recent seminar on "How to use AI to improve the efficiency of non clinical research under the FDA's new policies", attending experts stated that China's new drug research and development field has already begun to leverage AI. Improving prediction accuracy "AI can not only design molecular drugs, but also predict the probability of molecular drugs entering clinical practice and the possibility of market launch. ”Pei Jianfeng, a specially appointed researcher at Peking University's Institute of Frontier Interdisciplinary Studies, told Science and Technology Daily reporters. We continue to pay attention to the key issue of the success rate of new drug development. ”Pei Jianfeng said that the development of new drugs takes a long time, and 90% of new drugs will fail in clinical trials. Their team is developing AI technology to predict the success rate of various links in the long chain of new drug development, in order to guide R&D personnel to choose the right path in the "maze". Wang Shuhang, a researcher at the Clinical Trial Center of the Cancer Hospital of the Chinese Academy of Medical Sciences, introduced that the National Cancer Center has established a specialized institution to collect research data on rare diseases, new tumor targets, and multi omics. It continues to cooperate with AI companies, university teams, and other organizations to carry out clinical research guided by AI, and help more original drugs land. Discovering innovative drugs through target discovery is currently one of the main directions for new drug development. The star target PD-1 (programmed death receptor 1) has become a "lighthouse" in the development of cancer treatment drugs. Not all potential targets are suitable for drug development. If AI can predict the development value of different targets at the time of project initiation, developers will not have to wait until the third phase of clinical trials to discover that drugs have no efficacy, thus incurring huge trial and error costs. ”Pei Jianfeng introduced that in recent years, the AI technology developed by the team can provide optimal drug design solutions for multiple targets, helping to design small molecule drugs with higher drug efficacy. Zhao Yi, a researcher at the Institute of Computing Technology of the Chinese Academy of Sciences, said that the "Deep Generation Model PRnet" independently developed by their team, based on the neural network architecture and the generated adversary network, realized accurate prediction of the dynamic response of gene transcriptome under drug disturbance, and the prediction accuracy reached 87% in the validation of multiple disease models. Enhancing original ability "Clinical trials account for over 70% of the time and cost in new drug development, but it is difficult to provide accurate guidance for clinical decision-making. ”Professor Zhao Wei, Director of the Department of Clinical Pharmacy at the School of Pharmacy, Shandong University, believes that traditional clinical trials not only have long cycles and low efficiency, but also suffer from low accuracy and insufficient reliability in predicting therapeutic effects. For example, in terms of security assessment, the false positive rate should be at least 15%. The FDA commissioner also mentioned in the relevant notice that alternative methods can reduce research and development costs and drug prices, and provide safer treatment options for patients faster. Taking the key indicator "drug clearance" in safety assessment as an example, the traditional method predicts the highest effective rate of 65.8%. We trained the machine learning model based on research data from 35 medical institutions and universities, and improved the prediction efficiency to 94.1%. Zhao Wei said that when anti infective drugs such as amoxicillin and cefotaxime were used in newborns, the team used the model to predict doses with a clinical accuracy rate of 80%. Our country's ability to develop original drugs is increasing, but the lack of effective data from preclinical studies and the difficulty in climbing safety doses have become "hidden concerns" in the development of original drugs. New methods such as AI can accurately estimate many parameters, including safe doses, greatly enhancing the research and development capabilities of original drugs. ”Zhao Wei said. In addition to enhancing the overall ability to develop new drugs, AI can also discover new features of old drugs or "rescue" failed drugs. When an old drug is discovered, it focuses on one target, but when it acts on the human body network, it can demonstrate new therapeutic effects by acting on other targets, and the safety of the old drug is better, "said Tang Yun, a professor at the School of Pharmacy, East China University of Science and Technology. With the support of algorithms, Chinese research teams continue to explore new associations between drugs, diseases, and targets, and constantly enrich Chinese practices, providing effective innovative paths for revitalizing existing drug resources. (New Society)