New Drug R & D often takes a lot of time and money: it takes more than a decade to bring a new drug to market, and the cost of drug R & D has soared in recent years. A new analysis published in the Journal of Drug Discovery Today shows that the average cost of a new drug for Biopharma giants has reached $6.16 billion [1] . The waves of time and money always disperse with the tide, and the large investment of Biopharma companies in research and development may not bring corresponding returns.
How to avoid the situation that the return on investment is not proportional?
The whole pharmaceutical industry is exploring feasible solutions:
1) Shift to data-centric drug discovery with the help of artificial intelligence (AI);
2) The change of commercialization mode, more inclined to external cooperation and diversified pipeline layout;
3) Development of new modalities, such as macromolecules, cellular gene therapy, etc. [2] .
AI is slowly gathering momentum until it penetrates vertically into all the gaps in drug research and development. You and I are experiencing the transition of science and technology.
In terms of cutting-edge technology, AlphaFold2 won the first place [3] in the 14th Protein Structure Prediction Competition (CASP14) in 2021, showing its strong ability to predict protein structure and completely igniting the flame of AI drug research and development.
At the end of 2022, OpenAI released ChatGPT, followed by Google, which launched the Bard model. The big factories in Silicon Valley are at daggers drawn, and China is not to be outdone. Within a month, Ali launched Tongyi Qianwen, Baidu launched Wenxin Yiyan, Huawei’s big model is called Pangu, and Kunlun Wanwei’s big model is called Tiangong. At the beginning of chaos, AI is born.
Without warning horns and brakes, protein structure prediction and generative AI are like two carriages running side by side, colliding with the times.
In terms of industry, MNC has laid out AI Biopharma, the amount of investment and transactions has been rising, and the AI Biopharma track has been attracting money. In the six years from 2015 to 2021, the amount of Biopharma financing in AI increased nearly 20 times ( $386 million vs $7.504 billion), of which the amount of financing in 2020 quadrupled compared with the previous year ( $1.082 billion vs $4.443 billion). Even in 2022, when the industry suffered a cold winter, the financing events of the AI Biopharma track increased compared with the previous year.
Pharmaceutical transactions on the AI Biopharma track burst out in 2019, with the total amount of transactions exceeding $4 billion in the following year, and nearly quadrupled in 2021 compared with the previous year ( $4.075 billion vs $14.339 billion). The wave of capital is turbulent, and the only constant is the pursuit of innovative technology.
In fact, after the development of the past 10 years or so, AI has been widely used in drug research and development. From the point of view of research and development, AI can not only be used in pre-clinical processes such as target discovery and molecular design, but also improve efficiency and accuracy in clinical trials and shorten the trial cycle. From the perspective of drug types, AI has expanded from traditional small molecule drugs to macromolecule drugs, cell therapy, gene therapy, nucleic acid drugs and other fields.
With the rapid development of protein structure prediction and other technologies, the development of AI + macromolecular drugs has become more and more mature, and more and more companies have joined the track, including Fapon Digiwiser, which uses the unique “conformational selection mechanism” as the drug target theory.
Troika, heading for the future of AI and Biopharma.
Truth existed before the birth of man. It is found in the spark of wood rubbing, the apple falling from the tree, the steam from the kettle, and in the translation and folding of proteins.
In 1894, Emil Fischer, a German chemist, proposed the lock and key model, which holds that the structure of enzyme and substrate at their binding sites should be strictly matched, just as a lock matches its original key in structure. In 1958, Daniel E. Koshland revised the above theory and put forward the theory of induced fit, which holds that the active site of enzymes has a certain flexibility, and its conformation may change under the induction of substrates to achieve the greatest degree of fit.
However, the premise of the above two theories is that the structure and conformation of proteins are single and stable. The reality is that proteins have active and inactive conformations. The charm of science is that it is always iterating, bringing the public closer to the truth.
In 1998, Dr. Ma Buyong ended his postdoctoral career at the University of Georgia and became a researcher at the National Cancer Institute (NIH). After in-depth study of protein dynamic conformation and functional regulation, he summarized and put forward the conformational selection theory of biomolecular recognition, thus establishing three major theories in the field of substrate binding, namely, lock-key model, induced fit theory and conformational selection.
“Conformational selection mechanism” drug target theory refers to the design and optimization of drugs based on in-depth understanding and analysis of drug targets and drug conformational changes, combined with biological activity regulation. This theory is universal and can guide the design of small molecule drugs (such as allosteric drugs), macromolecular antibody drugs and nucleic acid drugs, which is a major shift from structure-based drug design to dynamic conformation-based drug design. Dr. Ma Buyong is currently a professor at the School of Pharmacy, Shanghai Jiaotong University, and the founder and chief scientist of Fapon Digiwiser.
Molecular simulation technology is based on computer simulation to calculate the physical, chemical and biological properties of molecules, which can directly obtain the conformation (three-dimensional structure), energy, dynamics and interaction mode of molecules, and directly explain the interaction between drugs and proteins, the mechanism of chemical reactions, and the characteristics of protein folding.
AI technology can predict the properties and characteristics of a series of molecules, transform or generate molecules, and accelerate the process of drug development by analyzing a large amount of data of biomolecules.
The technical solution of molecular simulation + AI covers a series of artificial intelligence design and prediction, including the dynamic conformation selection of drug receptors and ligands, the dual allosteric effect of biological drugs, the design and optimization of antibody drugs, and the regulation and design of efficient polypeptide molecular conformation selection.
A workman who wants to do his work well must first sharpen his tools. The research team of Fapon Digiwiser combines molecular simulation technology with AI technology, and uses the powerful computing power of modern high-performance computers to accurately calculate the conformational changes and physical and chemical effects of target protein/drug molecules. Through protein data acquisition, artificial intelligence and molecular simulation, the dynamic and correlation of atom-molecule-functional domains are accurately grasped, and a series of AI drug design and prediction functions are realized, such as dynamic conformation selection of drug receptors and ligands, dual allosterism of biological drugs, design and optimization of antibody drugs, and efficient regulation and design of polypeptide molecular conformation selection.
When the dominant concept of “conformational selection theory” collides with the dominant technology of “molecular simulation + AI”, a hot spark is created, which illuminates the four technological platforms of Fapon Digiwiser:
Protease design platform: The deep learning model is used to learn the evolutionary relationship of protein sequences from the complex multi-dimensional amino acid sequence space, break through the chemical space limitations of traditional drug design, and provide downstream sequence variants with natural physical characteristics, high diversity and pipeline screening conditions.
Virtual antibody screening platform: The deep learning model based on AI + molecular simulation training is used to realize the functions of antibody structure prediction, antibody affinity evaluation, antibody thermal stability/druggability/specificity evaluation and optimization.
Protein engineering platform: The existing protein stability data and the supplementary molecular dynamics simulation data are used for protein stability prediction and optimization, thermal stability prediction and other modules.
New drug and small molecule platform: use AI + molecular simulation to generate new small molecules based on multiple dynamic conformational target pockets. Can be use for that design of small molecule drugs, such as PROTAC, molecular glue and the like.
With the “troika” of advanced theory, technology and platform, Fapon Digiwiser takes self-developed computing platform as the research and development core, and gradually establishes the design module of drug function for macromolecules and small molecules, covering drug target discovery, biochemical drug molecular screening, protein and nucleic acid engineering, drug safety prediction, etc.
The company is committed to building a leading technical team of AI molecular design technology, becoming a problem solver of drug research and development, and establishing a leading AI for Science research and development model in the future, forming a mature R & D operation model of AI Biopharma. Science will open the unclosed door of history and shine through the eternal night with its own light. What Fapon has to do Fapon Digiwiser is probably to grow up and become one of the brilliant stars in the process of waiting.
A single thread does not make a thread, and a single tree does not make a forest. As a subsidiary of Fapon, Fapon Digiwiser also assumes the responsibility of enabling the integration of diagnosis and treatment for the group. Fapon Digiwiser and Fapon’s internal pipelines (Biopharma, diagnosis and treatment departments) cooperate in depth to undertake internal needs, apply AI + molecular simulation to internal pipelines, and polish and iterate AI products on pipelines, so as to establish a combination of dry and wet, algorithms, molecular simulation and wet experiments within the group. An integrated computer-assisted drug/reagent solution.
The road is blocked and long, and the line is coming.
The development of anything has to go through a tortuous process, and the AI Biopharma is bound to face stormy challenges from budding to growing into a towering tree.
On the one hand, data quality and quantity are still speed limiting steps and barriers. Data is the basis of AI model, but many data sources in drug development are different, often with noise, missing and inconsistency, which affect the reliability and accuracy of the model. For new targets and new molecules, the data in drug research and development are often limited, which also limits the development of AI models.
On the other hand, the boundaries of applicability of the model (interpretability and generalizability) and fit with wet experiments still need to be explored. What kind of problems AI model can solve in drug research and development, which kind of problems it is better at solving, and the applicability of the same model to other targets/molecules still need more exploration. From the perspective of drug R & D process, the proportion of AI optimization is relatively small, which does not break the traditional R & D system of Biopharma.
For the above challenges, on the one hand, Fapon Digiwiser can make up for some of the missing data by combining molecular and simulation methods; On the other hand, according to the advantages of Fapon in experimental ability, it constantly explores the new use scenarios of AI Biopharma in the research and development system of Biopharma, which can help to explore the new boundaries of the industry while achieving its own development.
Some people think that the Biopharma of AI has passed the financing outlet, coupled with economic factors, the investment of biological Biopharma tends to be more cautious and pragmatic, focusing on landing scenarios and commercial prospects. However, big language models such as ChatGPT provide incentives for the landing application of AI and promote the investment of related technologies.
How to dominate the ups and downs in the vast land? Fapon Digiwiser said that the company will continue to focus on technology, follow up the development of industry technology, explore the integration of molecular dynamics and AI technology path, and form technological differentiation; It also closely combines downstream experiments and Biopharma pipelines to polish effective and reliable drug calculation tools.
In the future, downstream AI Biopharma applications based on large models will emerge in endlessly, and new technologies in the fields of natural language, image and automatic driving will be continuously applied to the research and development of large and small molecule drugs. AI Biopharma will find a breakthrough in commercialization in the combination of production and research, dry and wet experiments, and get molecules that really surpass traditional drug design through calculation.
The past has not gone, but the future has come. The road is blocked and long, and the line is coming. Even if there are all kinds of dangers and obstacles, as long as we do not stop, the Biopharma of AI will eventually have a promising future.
References:
[1]Alexander Schuhmacher et al. Analysis of pharma R&D productivity - a new perspective needed. Drug Discov Today（2023）
[2]https://www.deep-pharma.tech/ai-in-dd-q3-2023
[3]https://www.predictioncenter.org/casp14/index.cgi
Source: Pharmacube.