Not long ago, the National Health Commission, the State Administration of Traditional Chinese Medicine, and the National Center for Disease Control and Prevention jointly released the "Reference Guidelines for Artificial Intelligence Application Scenarios in the Health Industry", listing 84 specific application scenarios including drug research and development, pressing the "fast forward button" to empower drug research and development with artificial intelligence (AI) technology. For a long time, there has been a famous "Double Ten" curse in the pharmaceutical research and development industry, which means that new drug development takes 10 years and 1 billion US dollars. How to break this curse, AI is highly anticipated. At the academic conference held recently, Chen Kaixian, an academician of the CAS Member, said that AI would bring $555 billion in value to the semiconductor industry and $1.2 trillion in value to the pharmaceutical industry. In recent years, generative AI has continued to make efforts in vertical industries. What does it bring to the development of new drugs? The reporter learned that the application and platform construction of a series of AI technologies are continuously improving the efficiency of drug research and development, and artificial intelligence technology may trigger a disruptive change in the pharmaceutical industry. Improving research and development efficiency plays a role in the entire drug development chain. By the end of 2023, the new antibiotic discovery problem that has plagued the industry for 60 years will be solved by AI. The article published in the journal Nature describes the process of scientists using AI technology to discover for the first time a new antibiotic against methicillin-resistant Staphylococcus aureus (MRSA). Using the antibacterial activity data of 39000 compounds against MRSA as a training "script", the MIT research team obtained an evaluation and prediction model for antibacterial ability. Subsequently, based on three deep learning models, the team "molded" an "appraiser" of compound human cytotoxicity, "screened" 12 million compounds, and finally obtained compounds that can fight MRSA and are safe for human health. This kind of experiment is almost impossible to complete solely by manpower. However, with its "hardcore" capabilities, AI has greatly reduced the time required to evaluate and verify each molecule. In the past, the success rate of designing drugs targeting specific targets was very low. Data shows that even the original targets reported in authoritative journals such as Cell and Nature have a drug success rate of less than 10%. Why is it still difficult to design drugs when the target is present? Taking small molecule chemical drugs as an example, based on the principle of "key unlocking", compounds are designed for the target. The candidate molecules in the compound library can reach hundreds or thousands, and screening is time-consuming and labor-intensive. The practice of new drug research has shown that it is becoming increasingly difficult to find better new drugs on old targets Chen Kai said that at the same time, the difficulty of discovering new targets is also increasing, requiring new ideas and technologies to "break through the situation". Artificial intelligence can provide unprecedented assistance to humans in discovering and predicting new targets. Chen Kai first introduced that foreign research data shows that the application of artificial intelligence technology can shorten drug design time by 70% and increase drug design success rate by 10 times. In theory, AI can play a role in the entire chain of drug development Chen Kai first told reporters that in the entire chain of new drug development, the discovery of a new target often leads to the production of a batch of new drugs and promotes breakthroughs in clinical treatment. China particularly needs to seize the opportunity period of AI assisted original pharmaceutical development Chen Kaixian stated that relevant data shows that in the past decade, the number of potential drug targets discovered in China each year has remained below 6, and new drug research in China is still mainly focused on catching up and following. In recent years, artificial intelligence technology has become a powerful tool for discovering new targets. For example, research teams such as Zheng Mingyue from the Chinese Academy of Sciences Shanghai Institute of Materia Medica have developed a new algorithm of "Facebook recognition" to identify new targets through correlation and comparison by extracting chemical structure characteristics, gene change characteristics, and drug activity characteristics. This technology has been successfully used to search for immune targets of the anti-tumor drug methotrexate. AI also has advantages in discovering new targets from literature knowledge. Chen Kaixian stated that a large amount of data has been accumulated from previous basic and clinical studies, and these findings are scattered and stored in research literature "unrelated" to each other, making it difficult for humans to discover the potential correlations between them. Artificial intelligence has powerful and efficient learning and analysis capabilities, which can mine the correlation relationships scattered in a large number of literature and promote the identification of new mechanisms and targets. Although there are currently no AI assisted new drugs approved for market in China, many new drugs have rapidly entered the clinical trial stage with the assistance of AI Chen Kai said first. By the end of 2024, research teams such as Stanford University published an article in the journal Cell stating that large-scale neural network models with multi-scale and multimodal capabilities have the ability to represent and simulate the behavior of molecules, cells, and tissues in different states. On this basis, AI virtual cells have gained credibility in high fidelity simulation, accelerated discovery, and guided research. Previously, the Proceedings of the National Academy of Sciences published an article stating that researchers have replaced carbon based "patients" in life forms with silicon-based "patients", and the simulation results are highly consistent with real data. In the study, 1635 "virtual patients" who "live in the computer" have suffered from breast cancer and cancer cells have metastasized. Through the experiment, the researchers found the optimal path for biomarkers to guide the clinical treatment of breast cancer. Based on in vitro, in vivo, clinical, population level, and multi omics data, researchers perform "digital twins" on patients' drug responses, generating rich pharmacological data of "virtual patients" for testing biomarkers, drugs, and other aspects. The imagination and thinking of human life activities can be transmitted to computing power in the form of data, which is the foundation for realizing virtual life or cells Xi Jianzhong, Vice Dean of the School of Future Technology at Peking University, told Science and Technology Daily reporters that in the half century development process, molecular biology has "interpreted" life through different levels of omics data, such as genomics, proteomics, transcriptomics, etc., accumulating a large amount of life science data. With the development of technological integration, human data acquisition capabilities are becoming increasingly strong. Optical imaging technology has now reached the nanoscale, allowing for the dynamic 'recording' of organelles in cells, "said Xi Jianzhong. A large amount of new data and research have given rise to new disciplines such as imaging omics. These breakthroughs not only deepen the interpretation of life, but also become the foundation of digital life. In fact, China's scientific research teams have already made arrangements in infrastructure, research topics, and other aspects. For example, in Huairou, Beijing, a multi-modal cross scale biomedical imaging facility that costs billions of yuan has begun to take shape. The cell imaging building, medical imaging building, and full-scale integration center form a powerful technological "aircraft carrier" with hard power. Among them, the full-scale data processing center will provide powerful computing support for related research. Different teams are working on digitizing some key organs. We hope to achieve digitalization of tumor cells Xi Jianzhong stated that tumors have high heterogeneity and dynamism, and everyone is different and constantly changing, making effective drug screening very difficult. Virtual tumor cells can tell us how internal signaling pathways change under the action of a certain drug Xi Jianzhong said that to achieve such a goal, it is necessary to build the "prototype" of tumor cells based on existing data and basic models, and then train it. In real life, drug trials need to be conducted. If a patient takes one medicine, at least thousands of patients are needed to test thousands of different medicines. In this way, implementation is difficult and costly Xi Jianzhong said that virtual cells can simultaneously "eat" thousands of drugs and obtain thousands of sets of data in one model, achieving high throughput and high fidelity, which will greatly improve the screening efficiency of tumor drugs. The most fascinating aspect of generative artificial intelligence for researchers is its' unexpected 'performance. Xi Jianzhong stated that in terms of scientific exploration, AI can break through the boundary limitations of different fields. For example, cross disciplinary studies on cardiovascular and infectious diseases may reveal surprises such as the potential of antiviral drugs to lower blood pressure. Shortening the research and development cycle to tackle the "hard bone" of rare disease drug development. In the field of drug development, rare disease drug development is a difficult "hard bone" to tackle. That's why the drug review and approval process has set up a dedicated "orphan drug" green channel for it. The average cost cycle for diagnosing rare diseases is currently 4 to 5 years. The shortage of patients has created a dilemma of "cooking without rice" - rare diseases are difficult to "see", and the small number of clinical trial cases is one of the challenges in the development of rare disease drugs. Why does it take so long to diagnose a disease despite the development of modern medicine? Rare diseases do not come with a 'nameplate' Liang Lungang, the head of BGI Genomics' AI project, told reporters that it is often treated as a common disease, and when a child's development is significantly lagging behind that of their peers, they often seek advice from the nutrition department. The diagnosis of rare diseases faces the problem of both ends being 'open' in terms of symptoms and genetic variations, while diagnosis requires various methods to achieve 'convergence' at both ends and ultimately obtain a matching 'connection' Liang Lungang said that patients may only be diagnosed with rare diseases after various trial and error attempts have failed to find the cause. Nowadays, as artificial intelligence enters the stage of generative big models and can answer various open-ended questions, rare disease diagnosis is expected to skip the "trial and error" stage. Recently, Zhang Shuyang, the director of Peking Union Medical College Hospital, introduced at the National Health Commission's pharmaceutical technology innovation conference that the first rare patient artificial intelligence large model GeneT has been launched to assist grassroots doctors in the diagnosis and treatment of rare diseases. From answering true or false questions like 'yes' or' no ', to being able to answer complex question and answer questions like' what is this and why ', AI has improved its diagnostic capabilities for rare diseases. Like other application fields, AI first quickly grasps professional knowledge in the field of rare diseases Liang Lungang introduced that publicly available rare disease datasets and literature, as well as data from BGI's testing services, will be transformed into AI "knowledge". Most importantly, BGI Genomics has collaborated with Peking Union Medical College Hospital to timely apply frontline experience in the clinical diagnosis and treatment of rare diseases, enabling AI to have clinical "experience". We not only inputted massive knowledge of rare diseases into AI, but also taught it how to think like a genetic expert Liang Lungang introduced that BGI Genomics has transformed the expert's thinking process into language that AI can understand, allowing the new model GeneT to learn how to accurately screen for genetic variations that cause rare diseases. The efficiency has been improved by 20 times, and the accuracy in simulated and real cases has reached 99% and 98%, respectively. Liang Lungang stated that after completing preliminary analysis, the final diagnosis of GeneT still needs to be reviewed by experts. Data shows that with the help of artificial intelligence models, the diagnosis time for rare disease patients is expected to be shortened from several years to less than 4 weeks, which has loosened the "iceberg" of the vast majority of rare diseases without specific drugs. Data shows that the number of rare disease drugs under development in China has significantly increased from 2017 to 2022, with an average annual growth rate of 34%. However, a study in the Chinese Journal of Clinical Pharmacy showed that approximately 43.9% of rare disease drug clinical trials had actual enrollments smaller than the target enrollments. With the support of the National Rare Disease Registry System, clinical cohorts for rare diseases have been established to promote drug development in related fields. This allows rare disease patients to receive early diagnosis, "said Liang Lungang." Seeing rare diseases will alleviate the problem of scarce clinical queues for rare disease drug development and provide strong support for rare disease drug development. Experts believe that in the next 3-5 years, China will enter a stage of rapid development in AI drug research and development. Artificial intelligence technology will shoulder the responsibilities of molecular optimization, synthesis route design, as well as automatic generation and analysis