Recently, three high school students detonated the medical circle. They used artificial intelligence (AI) engine for target discovery and identified a new therapeutic target for glioblastoma multiforme (GBM), glioblastoma multiforme GBM is the most aggressive and common type of malignant brain tumor, accounting for 16% of all primary brain tumors. The paper they co-authored was published in the international scientific journal "Aging" on April 26, showing the broad prospect of artificial intelligence system-assisted new drug development.
Artificial intelligence (AI) technology is reshaping the pharmaceutical industry in terms of cost and efficiency. In recent years, relevant experts believe that AI pharmaceuticals will become an opportunity for overtaking in the domestic pharmaceutical industry. AI pharmaceuticals should be used as an entry point to strengthen forward-looking policy support for this emerging field and promote the original and independent innovation of the entire Chinese innovative pharmaceutical industry. Maybe In the field of pharmaceuticals, the real "smart" creation in China lies in the mature application of artificial intelligence technology.
At present, in the field of pharmaceuticals, AI pharmaceuticals are mainly composed of three types of companies that cooperate with each other to jointly promote the drug research and development process, IT technology companies, drug research and development CRO companies, and large pharmaceutical companies. IT companies use their own Internet foundation and platform advantages to empower industry applications, while large pharmaceutical companies have relevant data on drug research and development, mature R&D pipelines, and senior drug experts. With complementary advantages, AI pharmaceuticals can already be seen Many mature cases.
(Summary of AI pharmaceutical cooperation published by some foreign-funded pharmaceutical companies)
China's AI pharmaceuticals started a little later than Europe and the United States, but they are developing rapidly and have more advantages in data and algorithms . The doctor talked about how he developed the road of AI pharmaceuticals of CDK4/6 inhibitors for breast cancer.
According to data released by the World Health Organization, as early as 2020, breast cancer will surpass lung cancer in new cases worldwide, becoming the world's largest cancer. my country is the country with the largest number of breast cancer cases in the world, with nearly 420,000 new cases every year.
CDK4/6 inhibitors are currently the best-selling drugs for the treatment of breast cancer worldwide, and are used to treat HR-positive/HER2-negative breast cancer patients. For example, Hengrui Medicine’s Darciclib, Pfizer’s Palbociclib, and Eli Lilly’s Abecicil have been approved for marketing in China, and Abecicil has also entered the National Medical Insurance List.
However, while CDK4/6 inhibitors bring good news to the majority of patients, there are also "flaws", such as the unavoidable occurrence of varying degrees of drug resistance and clinical side effects, and the competition for homogeneity is extremely fierce. This requires a unique approach to develop CDK4/6 inhibitor breast cancer drugs with a new mechanism.
In this regard, Wang Ziyi's team proposed that the kinase activity of CDK4/6 must rely on the key characteristic of itself forming a complex with CCND (cyclin D), and developed a protein-protein protein that can block CDK4/6-CCND interaction The small molecular compound can also inactivate CDK4/6 kinase, thereby achieving the purpose of inhibiting the growth of breast cancer cells.
After the idea of new drug research and development was determined, a long-standing problem emerged again, that is, the need to find the corresponding target molecule. The speed of this process, in the past, could only depend on luck. However, Wang Ziyi's team refused to stick to the rules, and chose a new path, using Wenxin's large-scale biological computing capabilities provided by the Baidu Flying Paddle Propeller Biocomputing Platform to carry out drug discovery work. The results amazed him, and within a few hours , 110 potential candidate molecules with high scores were screened out of a virtual screening library of 7.8 million compounds.
Lidebeck built a specific detection method to detect the activity of the screened compounds, purchased 40 of the 110 molecules for wet experiment detection, and finally found 6 high-potential molecules, of which 3 compounds can be activated at the same time To break the CDK4/6-CCND protein-protein interaction, there are 3 compounds that can break the CDK4-CCND protein-protein interaction. At present, the teams of the two parties are doing further research on these compounds, and it is expected to push this new type of inhibitor into the clinic in the near future.
According to reports, compared with the existing CDK4/6 inhibitors, the new drug is a first-in-class drug with an innovative mechanism, which has better specificity, and has more advantages in drug resistance and potential side effects. Bring the gospel. Thanks to these advantages, new drugs are expected to open up considerable market space.
Dr. Wang Ziyi said, "Baidu has domestic leading technical capabilities in AI+ drug research and development, especially Wenxin's large-scale biological computing model is very leading in China. Using these technical capabilities, Flying Paddle Propeller helps us find symptoms more efficiently . compound molecules, which greatly improves the efficiency of our drug discovery."
Baidu Paddle Helix ( Paddle Helix) is an "AI+biological" computing platform based on the deep learning framework of Paddle Helix. It provides Wenxin large model-biological computing large model capabilities. Multiple algorithm models have been opened, covering small molecule drug screening, peptides /Protein drug design, mRNA vaccine/drug design and other technologies, for new drug research and development, vaccine design, precision medicine and other scenarios, provide comprehensive algorithms for innovative pharmaceutical companies, medical technology providers, scientific research institutions, biotechnology companies, etc. in the field of biomedicine Tools and technical solutions.
At present, in the East China region, many companies have used the flying paddle propeller platform to carry out related drug research and development. In addition to Hangzhou Lidebaike and Baidu Flying Paddle to develop innovative breast cancer drugs, Suozhi Biology is also cooperating with Flying Paddle in many fields. Among them, HelixADMET, a large model for predicting ADMET properties, has been integrated into Sozhi's own AI drug The discovery platform (AIxMol®) was successfully applied to its pipeline projects under research, which effectively helped Suo Zhi to improve the success rate of synthetic test wet experiments, thereby improving the overall R&D efficiency, and successfully identified 3 PCCs in just 18 months molecular.
Data show that 40-45% of clinical trials fail due to high toxicity and low drug-like properties of drugs. Being able to rule out poorly behaved molecules early in drug development can save a lot of time and money. Therefore, compound druggability prediction (ADMET for short) is crucial to the success of new drug development. ADMET is the collective term for the absorption, distribution, metabolism and excretion behavior (ADME) and toxicity (Toxicity) of compounds in the body concerned by pharmacokinetics, and is the most important reference index for measuring the druggability of compounds.
To solve this problem, the industry has proposed many solutions, such as admetSAR, ADMETlab, swissADME, etc., but the data sets used for training of these methods are generally small in magnitude, so they are less effective in predicting the properties of compounds with unknown skeleton structures , At the same time, it is impossible to expand indicators based on user needs. The property prediction of compounds with unknown skeleton structure has poor generalization ability, the cost of drug design is high, and there is a certain resistance to scalability.
The HelixADMET large model of the flying propeller can calculate the ADMET related indicators of 1000 molecules within 60 seconds. Compared with many well-known ADMET forecasting software at home and abroad, it greatly surpasses competing products in terms of completeness of functions (prediction of 52 indicators) and accuracy of indicators (more than 4 percentage points higher than other comparison platforms). The corresponding research has been included in the top journal Bioinformatics in the field of bioinformatics.
It is foreseeable that in the field of biopharmaceuticals, the HelixADMET large-scale model can be applied to the compound optimization/screening stage to assist in decision-making on compounds that are prioritized for clinical use, avoiding possible risks in the later stage; it can also guide the formulation of academic/project research plans and reduce the probability of blind experiments ; At the same time, the model can be used to verify the drugability of new drugs/generic drugs, assess the risks of new drugs/generic drugs, greatly improve the efficiency of drug efficacy research, and evaluate and verify the effects of new drugs/generic drugs more quickly.
In addition, on May 2, the top international academic journal "Nature" published a breakthrough achievement in the field of biocomputing by Baidu and its partners - "Algorithm for Optimized mRNA Design Improves Stability and Immunogenicity", proposing an mRNA sequence optimization algorithm LinearDesign, Baidu impressively signed the research as the first completion unit. Taking the Spike protein of the new coronavirus as an example, the algorithm can find the most stable mRNA candidate sequence in just 11 minutes.
Experimental data proves that the LinearDesign algorithm design sequence will help biopharmaceutical companies quickly develop more effective mRNA vaccines, shorten the development cycle, and reduce R&D costs. The effectiveness of this algorithm has been verified in two vaccines, the new crown mRNA vaccine and the herpes zoster mRNA vaccine. Baidu's design significantly improved mRNA half-life and protein expression in vitro, and enhanced antibody responses in vivo by up to 128-fold compared to traditional benchmarks.
epilogue
The most widely used field of artificial intelligence is in drug research. In the development of a new drug, researchers often need to design, synthesize and evaluate multiple compounds to create potential new drugs. The process of refining promising compounds into drug candidates is often Both expensive and time consuming. If artificial intelligence is combined with drug research, not only the investment capital will be greatly reduced, but the efficiency will also be greatly improved.
However, for now, we are still facing a bottleneck in development. Whether AI pharmaceuticals can develop rapidly in the future should fully stimulate the vitality of my country's AI pharmaceutical industry from a systematic perspective, and provide support from multiple perspectives such as talent training, regulatory approval, park construction, and data management. Promote AI pharmaceuticals to realize the "revolution" of innovative drug research and development in my country, and finally let us look forward to the "fourth scientific and technological revolution" of artificial intelligence technology.