This article is organized by “Hua Xing has an Alpha” (ID: hxalpha). The keynote speech of the 4th “Influence Investment Summit” hosted by Lai Lipeng at Huaxing Alpha in 2019.Lai Lipeng, co-founder of Jingtai Technology, believes that AI can help pharmaceutical chemists break through the limitations of existing experience.Based on deep learning, according to the specific needs of the project, flexible combination of different methods is a feasible way to effectively explore and screen drug molecules in 10 compounds of 60 possible compounds.Lai Lipeng, co-founder of Jingtai Technology, is the speech of the theme forum of “Biotechnology Innovation Drives Change” in the 4th “Influence Investment Summit” hosted by Huaxing Alpha in the 4th “Influence Investment Summit” hosted by Huaxing Alpha.Starting from Jingtai Technology, we will share the practical experience in the research and development of AI+ new drugs from a technical perspective.The long road to new drug development: computing as the core solution for pharmaceuticals like investment, when you advance something, you need to let yourself believe that this thing can be pushed forward.New drug development usually mentions three “10”.It takes 10 years or more for a drug to go from initial research and development to final market. It costs more than 1 billion US dollars, but the success rate is less than 10%.In the past 20 years, the R&D investment of pharmaceutical companies has probably doubled, while the number of newly approved drugs has not changed accordingly.Scientists are very enterprising and have been exploring ways to accelerate drug development.These include DNA coding libraries, automated experiments, computations in quantum mechanics, molecular dynamics simulations, and artificial intelligence methods for special fires in recent years.And we think that cutting from a computational perspective is a more promising approach, and certainly not just by this single approach, it may be necessary to combine different approaches.In the practice of Jingtai Technology, we constructed two calculation methods at the top: First, calculation methods based on quantum physics, quantum chemistry or force field, including DFT, first-principles calculations, etc.As well as the simulation of molecular mechanics and molecular dynamics, we believe that there will be much room for improvement in the underlying technology.Second, combine the methods of deep learning.Using the method model of Deep Learning in practice, the results performed better in some systems.Both computing methods require a lot of computing resources behind them.The method of computational chemistry, like DFT, can be considered very accurate, but the barrier used is that its computational resources are expensive and time-consuming.Currently, relying on our excellent engineering team, we put the main computing resources in the cloud.When a large amount of computing resources are needed, it can be flexibly expanded to ensure the required amount of computing resources, and the computing nodes can be turned off when not needed, thereby saving the computing cost of the enterprise.The pharmaceutical industry is ultimately guided by experimental results.Regardless of whether you use Deep Learning or other machine learning methods, the customer ultimately values the test results.What can AI do?Answers from the field of drug research We have been thinking about what AI can do.We find that AI has a great value that can break through the limitations of human thinking framework.Medicinal chemists are based on their knowledge and past experience when designing drugs.Experience is certainly limited. The greatest value of AI in the field of drug research is to help drug chemists break through the limitations and provide new insights for drug chemists.An article in the 2017 “Nature” mentioned that the number of possible compounds is 10 to the 60th power, which means that in principle, the number of molecules we can use to make drugs is several times higher than the number of atoms in the solar system.Magnitude.The library we are currently using for physical screening may be 100,000, or millions.Now the DNA coding library can reach tens of billions, or hundreds of billions, but it is insignificant compared to the 60th power of 10.How to explore such a huge potential space?This requires very important technical means.In the field of medicinal chemistry, we believe that deep learning and machine learning can help us see a broader chemical space.Of course, we are not aimlessly exploring the 10th 60th space.Even if all the computing resources available on Earth are added together, there may be no way to deal with such a large amount of information.In the design process of the algorithm, there are many techniques that allow us to effectively search in this space.The specially designed molecular generation algorithm can not only effectively learn the existing molecular structure characteristics, randomly generate a large number of new molecular structures, but also can generate new molecular structures that meet the needs of a certain drug design according to the needs of pharmaceutical chemists, thereby greatly expandingHigh quality drug molecule candidates available during drug screening.Exploring the drug candidate space: Integrating and building a new drug development platform ID4 Now we have constructed such a chemical space, we believe that good drug candidates are in this space, the next question is what we use when we know their existenceKind of tools to pick them out.Combining these methods, including quantitative calculations, molecular dynamics simulation methods, and our own computational chemistry modules, as well as artificial intelligence methods, and the cloud computing technologies behind them, have built a platform for new drug development——ID4.The performance of the platform seems to be good at first. We are currently using this platform in two service directions: first, early drug research, including target finding, drug binding site prediction, and small molecules after protein targetDrug screening, screening for lead compound optimization, and for the prediction of crystal forms and later experimental studies.Compared with the traditional research and development methods, the platform will greatly save the energy of the research and development personnel, so that the work of professionals in medicine, chemistry and other fields will be concentrated at key nodes, thereby exerting greater value.In our actual cooperation project, with the ID4 platform, we can consider multiple attributes of candidate molecules in the screening process, including activity, drug-forming properties, novelty of the skeleton, synthesizable and so on.The results of the screening have also been initially confirmed by the experiment.Second, another important application of the platform is crystal prediction.Everyone usually takes the form of a tablet or a capsule. The peeling of the capsule is magnified, and each solid has a crystal structure inside.Different crystal structures are different three-dimensional stacking methods of small molecules, and the packing mode affects the physical properties of the crystal.Therefore, the medical field is very concerned about whether its crystal structure is the optimal structure.Whether a crystal transition occurs under certain conditions.There is no way to guarantee comprehensiveness by relying solely on experimental methods, and the calculation method is easier to do.The above is just a brief introduction, the technology is endless, but we are constantly advancing the boundaries of technology and scientific research applications.Disclaimer: This article represents only the guest position and does not represent the opinions of Huaxing Capital Group (“Huaxing Capital”).Huaxing Capital does not guarantee the accuracy or completeness of the article and is for your reference only.Huaxing Capital shall not be liable for any loss arising directly or indirectly or in connection with the use of the materials in this article..
Jingtai Technology Lai Lipeng: What can AI do in the development of new drugs?