Divamics, an AI-Powered Drug Discovery 'Innovation Factory' Focused on 'Undruggable Targets', Completes Angel Financing and Files IPO Prospectus

"Transforming 'difficult drug targets' into 'hot targets' represents the broadest application prospect and the most imaginative market space for AI technology in the field of drug development in the future. This is also Divamics' differentiated competitive strategy, positioning itself as an 'innovation factory' for new drug research and development."

AI Drug Discovery Company Divamics Completed Angel Round Financing in May, with Hangzhou Shileng Investment Management Co., Ltd. as the Investor. This Round of Financing Mainly Aims to Introduce Strategic Shareholders to Boost the Company's Business Development. Reportedly, the Company Has Launched a New Fundraising Plan, Which Will Be Mainly Used for Team Building and the Layout of New Innovative Drug Research and Development Pipelines.

Regarding Divamics, it is positioned as an interdisciplinary technology enterprise driven by molecular simulation and modeling techniques to advance new drug development. Its core technology is a molecular simulation computing platform, including a machine learning-based molecular force field engine and parallelized molecular dynamics simulation technology, aiming to design and optimize lead compounds based on the dynamic mechanisms of protein target structures and their interactions with drug molecules, providing a new entry point for drug research and development.

According to Dr. Zheng Zheng, founder of Divamics, international computer-aided drug discovery technologies are roughly divided intoPhysics methods based on first principles and statistical methods represented by AI algorithmsTwo technical schools. The physical model is a bottom-up analytical model, characterized by high accuracy, less reliance on known information, but with expensive computational costs. AI algorithms, on the other hand, only require inputting training data into the training model and waiting for the model to converge before application, significantly reducing development costs compared to the former. However, the issue lies in that the model's accuracy is limited by the quality of the training data and the scope of biological system diversity covered.

As a result, the development of these two types of algorithms exhibits completely different development models. The former (physical models) often has a long development cycle (several years), with relatively low requirements for algorithm iteration frequency after completion, but high development difficulty. Currently, there are only a few teams in China with the capability to develop physical models. The latter (AI models) is characterized by relatively lower development difficulty (currently, the variety of AI drug design software products globally is countless), but with high iteration frequency, requiring continuous introduction of new databases for updated training. However, their applicability for system expansion beyond the database is poor, heavily relying on the coverage of previous research systems.

To this end, Divamics leverages the development characteristics of AI models and their impact on algorithm development patterns. It introduces AI models into the physics engine in a targeted manner, replacing the most costly part of development—the molecular force field—with AI models. At the same time, by combining the conformational change features presented by different target systems during molecular dynamics simulations, Divamics uses AI models to pre-learn the binding sites and pathways between drugs and targets, thereby improving computational speed. In simple terms, by integrating AI models with physical models, it reduces the development cost of physical models while retaining their applicability across various biological systems, enhancing computational speed. This approach preserves the advantages of both methodologies: high precision, low data dependency, and fast computation.

Computer-aided drug design technology, after decades of development, has been deeply integrated into various processes of drug research and development, playing a crucial role in multiple stages primarily involving molecular-level drug design and analysis. Moreover, the industry's demand for computer-aided drug design technology has gradually expanded from ultra-large-scale virtual screening and lead compound optimization to areas such as exploring drug mechanisms of action and designing complex macromolecular drugs and their delivery systems. This not only places higher demands on the computational efficiency and accuracy of related algorithms but also presents significant challenges to the generalization capabilities of these algorithms when applied to unknown complex molecular systems.

To this end, Divamics has developed and integrated a drug molecule R&D platform through extensive computational simulation research and testing on various drug-target systems. The platform expands the focus from the binding interface of drugs and targets to the dynamic performance of target functional conformational changes after the binding of drug molecules.By comparing the kinetic behavior of cavity proteins without bound drugs and analyzing the thermodynamic properties during the binding process of drugs with their targets, a comprehensive evaluation and prediction of the in vitro activity of drug molecules can be performed.

Specifically, the computational process of its technical platform begins with structural modeling of the target and the restoration of dynamic conformations in physiological processes: by analyzing stable conformational states formed during the conformational changes of the target, potential drug-binding sites are selected for high-speed virtual screening algorithms and high-precision molecular simulation algorithms to perform hierarchical calculations, yielding reliable hit compounds; subsequently, through a combination of dry and wet experiments, the pharmacodynamic mechanisms at the molecular level are clarified, and ultimately, based on the principle of molecular scaffold hopping, novel lead compound molecular structures are designed.

Based on this technology, a key future development goal for Divamics is to expand into the market of "difficult-to-drug targets," which account for a larger proportion of known targets. The company aims to leverage its technical advantages to accurately and rapidly simulate the structure and physiological properties of these "difficult-to-drug targets," explore binding sites, and design drug molecules that can effectively bind to them. Dr. Zheng Zhen stated,"Transforming 'difficult-to-drug targets' into 'hot targets' represents the broadest application prospect and the most imaginative market space for AI technology in the field of drug development in the future. This is also the differentiated competitive strategy of Divamics."

It is understood that Divamics' model is benchmarked against Relay Therapeutics, an American AI pharmaceuticals listed company. Relay Therapeutics is a drug research and development company driven by computer-aided drug design, and Schrödinger is one of its major shareholders. Relay Therapeutics centers around Schrödinger's algorithm products, integrating other competitive algorithm products from the market and combining them with self-developed AI models to form a computing platform covering the preclinical stage of drug development. At the same time, it conducts drug research through a combination of computational and experimental methods. "This is highly similar to Divamics’ current and mid-to-long-term planned model."

Currently, the company has collaborated with multiple pharmaceutical enterprises both domestically and internationally, initiating joint research and development efforts on several drug pipelines. Within a year, it has propelled one drug into the clinical trial stage. Meanwhile, through the collaborative R&D model, it has advanced a drug pipeline related to an "undruggable target" associated disease into the patent application stage. In terms of commercialization, following the earlier collaboration with Nanjing Ruichu at the beginning of the year on three neurodegenerative disease-related joint R&D projects, the company has now added another multi-million-dollar joint R&D pipeline partnership. Additionally, it has reached an LNP technology co-development agreement with a certain CRO enterprise.

As is known to all, new drug development is an extremely huge market. According to statistics from EvaluatePharma, global pharmaceutical R&D investment reached 178.9 billion US dollars in 2019, with a compound annual growth rate of 4.64% from 2013 to 2019; it is expected to reach 213 billion US dollars by 2024, with a compound annual growth rate of 3.23% from 2019 to 2024.

In this market, the pipelines of most domestic AI pharmaceutical companies are still in the research and development stage, and only a few companies have advanced their AI drug pipelines to clinical trials. Dr. Zheng Zhen believes that the development costs for Me-too or FIC/BIC drug pipelines are similar, with contract prices for drug pipelines generally starting in the tens of millions. In the next five years, AI pharmaceutical companies are expected to expand the research space for drugs related to difficult-to-drug targets, with the market size for AI drug research and development projected to reach 10 billion yuan by 2025. "Although the number of pipelines that may achieve breakthroughs in the near term is limited, pipeline revenue remains considerable. Moreover, with the improvement in R&D efficiency and the completion of contractual 'milestone' tasks, future revenue is expected to double."