Not long ago, the National Health Commission, the state administration of traditional Chinese medicine, and the State Administration for Disease Control jointly issued a reference guideline on artificial intelligence application scenarios in the health industry, listing 84 specific application scenarios such as drug research and development, the“Fast-forward” button has been pressed to enable drug development through artificial intelligence (AI) technology.
There has long been a well-known“Double-ten” mantra in the drug development industry that new drugs take 10 years and $1bn to develop. There are high hopes for how to break the spell. At a recent academic conference, Chen Kaixian, a Chinese Academy of Sciences Academician, said AI would be worth $555bn to the semiconductor industry and $1,200 bn to the pharmaceutical industry.
In recent years, generative AI in the vertical industry continues to force. What does it bring to drug discovery? The reporter learned that a series of AI technology applications and platform construction are constantly improving the efficiency of drug research and development, artificial intelligence technology may lead to disruptive changes in the pharmaceutical industry.
Improve the efficiency of R & D
Play a role in the whole chain of drug research and development
In the 2023, AI solved a problem that had plagued the industry for 60 years: discovering new antibiotics. Nature describes how scientists used AI technology to discover for the first time a new antibiotic against methicillin-resistant Staphylococcus aureus (MRSA) .
Using the antimicrobial activity data of 39,000 compounds against MRSA as a training“Script”, the MIT research team obtained an evaluation prediction model for antimicrobial ability. Based on three deep-learning models, the team then“Shaped” the human cytotoxicity“Judger” of the compounds, “Screening” 12 million compounds, the result is a compound that fights MRSA and is safe for humans.
Such an experiment would be almost impossible to carry out by human hands alone. However, AI, with its“Hard core” capabilities, has greatly reduced the time it takes to evaluate and verify each molecule.
In the past, the success rate of target-specific drug design has been very low. Even the original targets reported in journals such as cell and nature have less than a 10 per cent chance of success, according to the data.
With targets in place, why is it so hard to design drugs? Taking small molecule chemical drugs as an example, according to the principle of“Key unlocking”, designing compounds for targets, there are hundreds of candidate molecules in the compound library, and screening is time-consuming and labor-intensive.
“The practice of new drug research shows that it is becoming more and more difficult to find better new drugs on old targets,” Chen said. At the same time, finding new targets is becoming more and more difficult, and new ideas and technologies are needed to“Break the ice”.
Artificial intelligence can provide unprecedented help for human beings in the discovery and prediction of new targets.
Chen Kaixian introduced, foreign research data show that the application of artificial intelligence technology can make drug design time shortened by 70% , drug design success rate increased by 10 times.
“In theory, AI can play a role in the whole chain of drug research and development,” Chen told reporters. In the whole chain of New Drug Research and development, the discovery of a new target often leads to a batch of new drugs, promote the breakthrough of clinical treatment.
“Our country especially needs to seize the opportunity for AI to help the original development of medicine,” Chen said, the number of potential drug targets found in our country every year has been hovering below 6, and the new drug research in our country is still dominated by catch-up and follow-up.
In recent years, artificial intelligence technology is becoming a powerful tool to discover new targets. For example, Zheng Mingyue and other researchers at the Chinese Academy of Sciences’ Shanghai Institute of Materia Medica have developed a new algorithm for“Face recognition”, by extracting the chemical structure characteristics, gene change characteristics, drug activity characteristics to identify new targets. The technique has been successfully used to find immune targets for older anti-cancer methotrexate.
Ai also has advantages in finding new targets from literature knowledge. Chen said large amounts of data had been accumulated from past basic and clinical studies, and the findings were stored“Uncorrelated” in the research literature, making it difficult for humans to discover potential links between them. Artificial intelligence has powerful and efficient learning and analysis capabilities, which can mine the correlation scattered in a large number of literatures and promote the identification of new mechanisms and new targets.
“Although no new drugs developed with the help of AI have been approved in our country, many new drugs have entered clinical trials with the help of AI,” Chen said.
Reduce the cost of research and development
Drug trials are no longer expensive
At the end of the 2024 year, a team of researchers at Stanford University and elsewhere reported in the journal cell, large-scale multi-scale and multi-modal neural network models have the ability to represent and simulate the behavior of molecules, cells and tissues in different states. On this basis, AI virtual cells have the credibility of high-fidelity simulation, accelerating discovery and guiding research.
Earlier, Proceedings of the National Academy of Sciences of the United States of America reported that researchers had replaced life-form carbon-based“Patients” with silicon-based ones, and that the simulation results were highly consistent with real data.
In the study, 1,635“Virtual patients” who were“Living on a computer” developed breast cancer that had metastasized. Through the trial, the researchers found an optimal path for biomarkers to guide clinical treatment of breast cancer.
Based on in vitro, in vivo, clinical, population-level, and multi-omics data, researchers perform“Digital twinning” on patients’ drug responses to generate rich pharmacodynamic and pharmacological data for“Virtual patients,” which can be used to analyze patients’ drug responses, for testing biomarkers, drugs, etc. .
“Human imagination and thinking about life activities can be transferred to computing power in the form of data, which is the basis for realizing virtual life or cells,” Xi Jianzhong, deputy dean of the School of future technology at Peking University, told Science and technology daily, in the past half century, molecular biology“Interprets” life through omics data at different levels, such as genomics, proteomics, transcriptomics, etc. , and has accumulated a large amount of life science data.
With the development of technology fusion, the ability of human beings to obtain data is getting stronger and stronger. “Optical imaging technology has now reached the nanoscale and Can ‘record’ the dynamics of organelles in cells,” said XI, adding that new disciplines such as imageomics have emerged from a wealth of new data and research. These breakthroughs in the in-depth interpretation of life, but also become the basis of digital life.
In fact, our country scientific research team already had the layout in the infrastructure, the scientific research topic and so on. In Huairou District, for example, a multimodal and multiscale biomedical imaging facility has begun to take shape at a cost of zero yuan, cell Imaging Building, medical imaging building, full-scale integration center and other scientific and technological“Aircraft carrier” with hard power, in which the full-scale data processing center will provide strong computing power support for related research.
“Different teams are working on digitising some key organs. We hope to digitise tumour cells,” says XI, tumor is highly heterogeneous and dynamic, so it is very difficult to screen effective drugs.
“The virtual tumor cells can tell us how the signaling pathways inside the cells change under the action of a certain drug,” XI said, the“Rudiment” of tumor cells needs to be constructed based on existing data and basic models, and then trained.
“In real life, drug trials are conducted. If one patient takes one drug, thousands of drugs need to be tested on at least thousands of patients. This makes it difficult and costly to carry out,” said XI, virtual cells can simultaneously“Eat” thousands of drugs and obtain thousands of sets of data in a set of models, achieving high-throughput and high-fidelity, which will greatly improve the efficiency of cancer drug screening.
What most fascinates researchers about generative AI is the element of surprise. When it comes to scientific exploration, AI can break down the boundaries of different fields, says XI. For example, cross-sectional studies of cardiovascular and infectious diseases may yield“Surprises” such as the antiviral drug’s potential to lower blood pressure.
Shorten the cycle of research and development
Rare disease drug research and development“Hard bone”
In the field of drug research and development, Rare Disease Drug Research and development is a hard nut to crack. Because of this, drug review and approval for it opened a special“Orphan Drugs” green channel.
4 to 5 years, which is the current average elapsed time for rare disease diagnosis. Rare diseases are difficult to“See”, and the small number of clinical trials is one of the difficulties in drug development for rare diseases.
With the development of modern medicine, why does it take so long to diagnose a disease?
“Rare diseases don’t come with a nameplate,” Liang Lungang, head of the AI program at BGI, told reporters. “It can be considered a common disease, such as when a child is significantly behind his peers.” Often consults with the nutrition department.
“Rare disease diagnosis is faced with the problem of ‘opening’ both ends of the spectrum of symptoms and genetic mutations,” Liang said. “Diagnosis is to achieve the ‘convergence’ of both ends through various methods, and finally get the ‘connection’ of a match.” Patients may be diagnosed with a rare disease only after various trials and errors have failed to find the cause. Now that artificial intelligence has moved into the generative large-scale model stage, where it can answer open-ended questions, rare disease diagnostics may be able to skip the“Trial and error” stage.
The first large-scale artificial intelligence model for rare diseases, Genet, has been launched to help primary Peking Union Medical College Hospital doctors diagnose and treat rare diseases, president Zhang Shuyang said at a press conference on Medical Science and technology innovation held by the National Health Commission.
From“Yes” or“No” questions to complex“What is this? Why?” Questions, AI has improved the diagnosis of rare diseases.
“Like other application fields, AI first quickly grasped the expertise in the field of rare diseases,” said Liang, who opened rare disease data sets and literature, and data from BDA’s testing services will be transformed into AI’s“Knowledge”. Most importantly, BGI is working with Peking Union Medical College Hospital to provide AI with clinical“Experience” through timely application of frontline experience in the clinical diagnosis and treatment of rare diseases.
“We have not only given AI a huge amount of knowledge about rare diseases, but also taught it how to think like a genetic expert,” Liang said, let the new model GeneT learn to accurately screen for genetic mutations that cause rare diseases, with a 20-fold increase in efficiency and accuracy rates of 99% and 98% in simulated and real cases, respectively.
Liang said that GeneT at present after the completion of preliminary analysis, the final diagnosis still needs expert checks.
Data show that with the help of artificial intelligence model, rare disease patients diagnosed time is expected to be shortened from a few years to less than 4 weeks, which makes the vast majority of rare diseases no specific drug“Iceberg” began to loosen.
Data show that the number of drugs under development for rare diseases in our country increased significantly from 2017 to 2022, with an average annual growth rate of 34% . However, a study in the Chinese Journal of Clinical Pharmacy showed that about 43.9% of rare disease drug trials actually enrolled fewer patients than the target enrollment.
With the support of the National Rare Disease Registry system, rare disease clinical cohorts have been established to promote drug research and development in related fields. “This will allow patients with rare diseases to be diagnosed early,” liang said, adding that“Being seen” would ease the scarcity of clinical cohorts for rare disease drug development and provide strong support for rare disease drug research and development.
Experts believe that in the next 3-5 years, our country will enter the stage of rapid development of AI drug research and development, artificial intelligence technology will be responsible for molecular optimization, synthetic route design and automatic generation, automatic analysis, automatic screening of the whole process of research.
“We expect that the first AI-designed drug will be approved as soon as possible. Translation and application will still be the key to achieving high-quality development in the pharmaceutical industry,” XI said.