Tencent's AI drug discovery platform breaks through the closed-loop capability of high-throughput dry and wet experiments

The era of artificial intelligence (AI) involvement in drug discovery is fast approaching. As a powerful engine in the field of new drug research and development, the application of AI in disease modeling, target identification, compound screening, drug design and clinical trial optimization is not only expected to greatly reduce the time, manpower and cost of new drug development, but also promote innovation in the field of drug discovery, discover targets and drug mechanisms that are otherwise impossible to discover, and create new drug assets and incremental markets.

"There are many domestic AIDD companies, and when we enter this track, we will naturally do something different." Recently, at the T-Inspire and 2022 Tencent Health Smart Pharmaceutical Open Day, Liu Wei, technical director of Tencent Health's AIDD, showed the new progress of Tencent's first AI-driven drug discovery platform and "something different". "Based on Tencent's long-term accumulation and algorithm innovation in deep graph learning, super computing power, big data capabilities and interdisciplinary capabilities, Tencent has empowered all steps of preclinical research in drug development with AI and formed an integrated end-to-end AIDD service."

Powerful deep graph learning ability to create a CAMEO and CASP double champion model

The emergence of new technologies has injected new vitality into the research and development of new drugs. There are two levels of AI technology to empower the field of new drug discovery, the first level, modeling from the perspective of microscopic and underlying physics, such as around the interaction of molecules and targets or prediction of the properties of molecules; The second level is to model the data correlation between compounds, proteins, genes, and diseases at the macro level. In the industry's view, in the field of life sciences, there must be original models and methods to truly solve a specific problem in the research and development process and create clinical value.

tFold, a new algorithm framework for protein structure prediction developed by Tencent's AI drug discovery platform, has proven its innovative value and effectiveness on CAMEO, an internationally recognized authoritative test platform, and has remained the weekly champion for several months. In the molecular generation based on graph learning, the skeleton transition molecular generation algorithm is used to discover nM-level lead compounds. Under the condition that the skeleton remains unchanged, the user can specify the retained structural part and iterate in the variable part, and finally realize the protection of the original molecule patent or optimize the ADMET properties of the molecule on the basis of ensuring the activity of the original drug molecule or lead compound.

The ADMET predictive model of Tencent's AI drug discovery platform can also demonstrate the importance of deep graph learning advantages. Liu Wei said at the meeting: "In the ADMET prediction cooperation we reached with pharmaceutical companies, Tencent AI's prediction function and internal data optimization can achieve a performance improvement of more than 30%, forming a positive cycle of testing, feedback and model iteration. At present, the AMDET prediction model has been well used in the study of chemical druggability optimization of small molecule medicines, and the prediction results on most attributes can reach more than 90% correlation. ”

The cross-domain combination of "AI+quantum chemistry" realizes high-precision quantitative calculation of drug-like molecules

The introduction of quantum mechanics at the end of the 19th century opened the door to explaining the microscopic material world, completely changing human understanding of the structure and interaction of matter. Quantum chemistry is the application of quantum mechanics to deeply understand the physicochemical properties of atomic and molecular systems. Traditional computational chemistry methods are difficult to balance the relationship between the size of the system and the amount of calculation, either the calculated system is particularly small, or the calculated system results are not accurate, and there is no way to ensure the accuracy in the case of a large system. The existence of this contradiction makes traditional methods unable to be applied to solve complex solutions, proteins and other problems.

With the accumulation of data and algorithm iteration, artificial intelligence is empowering and promoting the field of new drug research and development, combining AI technologies such as NLP large models, deep graph neural networks, and generative models with traditional pharmaceutical links, and comprehensively improving the efficiency of new drug research and development and expanding the technical application of drug innovation space through data cross-comparison, accelerated screening, and de novo generation. In order to meet the high accuracy of large-scale system computing, Tencent's AI drug discovery platform attempts to use the combination of "AI + quantum chemistry" to greatly reduce the amount of quantum chemistry by embedding the function of quantum chemical computing into the R&D pipeline based on artificial intelligence algorithms, and obtain a more accurate result than low-computing computational chemistry. Relying on Tencent's computing power advantage, Tencent's AI drug discovery platform basically does high-precision quantitative calculations of all drug-like molecules.

Super computing power and big data capabilities drive the closed-loop innovation of dry and wet experimental data of pharmaceutical companies

Embracing the general trend of AI pharmacy, more and more companies are actively laying out the "AI+pharmaceutical" track. However, AI drug research and development is still a data-driven state, and the breadth and quality of training data are very important to it, which makes AI pharmaceuticals have to face many difficulties such as large differences in data collection, uneven quality, and difficult data acquisition of failed results. For example, many companies often encounter such a problem in the application of AIDD: an AI algorithm made on the A target can well predict the molecules related to the A target, but if it is used on the B target, the final result is completely different, or even completely unavailable. There is a long way to go to overcome the problems, but on the other hand, these cutting-edge pain points will also become the core competitive points in the field of AI pharmaceuticals.

In order to solve this problem, Tencent AI drug discovery platform has made an out-of-distribution research framework DrugOOD, under which the existing databases are categorized, a large number of practical scenarios are divided, and the AI scoring system is used to evaluate the reliability of AI results generated between different targets, and early detection of the problem of mismatch between models and targets in follow-up research to optimize R&D efficiency.

At present, Tencent's AI drug discovery platform has reached cooperation with a number of pharmaceutical companies, and the prediction accuracy of the model has been verified in wet tests in a number of actual R&D scenarios. Based on the super computing power of drug screening cloud service, the screening speed and screening chemical structure space have been improved by orders of magnitude.

For the current pharmaceutical field, the deep integration of artificial intelligence technology and biomedical R&D and manufacturing will surely promote the transformation and upgrading of the biomedical industry and give birth to new formats and models. In the future, Tencent's AI drug discovery platform will further deepen industrial cooperation by providing advanced AI pharmaceutical algorithms, cloud computing super computing power, and industry-leading algorithm expert services, and jointly promote the drug research and development process with pharmaceutical companies through the combination of dry and wet laboratories.