news

한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina

Text｜Hu Xiangyun

Editor: Hai Ruojing

36Kr has learned exclusively that recently, the biomedical large model company "Shuimu Molecule" has completed nearly 100 million yuan in financing. Among them, the angel round was led by Huashan Capital, Daotong Investment and iFlytek Venture Capital participated in the investment; Qingzhi Capital participated in the seed round of financing. The raised funds will be mainly used for the development of biomedical multimodal large models and the conversational drug development assistant tool ChatDD.

Shuimu Molecular was incubated and established by the Institute of Intelligent Industry (AIR) of Tsinghua University in 2023. It mainly conducts basic large-scale model research in the biopharmaceutical industry and has developed a conversational drug development assistant tool ChatDD. Professor Guoqiang of Tsinghua University and Nie Zaiqing, chief researcher of AIR, serves as the company's chief scientist.

Academician Zhang Yaqin, Dean of the Institute of Intelligent Industries at Tsinghua University, said that AI+Life and Health is one of AIR's core research directions, and a series of research progress has been made, including accurate prediction of protein structure, AI antibody design, AI molecule design, etc. On this basis, the industry-university-research cooperation between AIR and Shuimu Molecular will form better assistance and synergy.

In Nie Zaiqing's view, human-machine collaborative drug development assistants are an inevitable trend in future drug research and development. In the past few years, although the application of AI technology has shown certain potential in drug discovery and optimization design, it also faces problems such as insufficient training data, single processing modality, and separation of information and knowledge. "The misunderstanding of AI pharmaceutical manufacturing at this stage is that it relies too much on the role of AI and hopes that the algorithm can directly generate candidate molecules or drugs. However, in the pharmaceutical manufacturing process, the experience and intuition of experts are often irreplaceable, so the best way is actually to combine the two."

The multimodal big model is the most likely way to achieve this goal, because compared with traditional AI pharmaceuticals, the big model has added a link to "align" natural language and biological coding language. It can be understood that each protein and molecule is a knowledge point that is related to each other, and while the model finds the connection between knowledge points, it can also integrate the experience of drug researchers through text questions, constantly "stimulating the thinking of people and big models in both directions" to find a better solution.

To achieve this goal, Shuimu Molecular first developed the GhatDD-FM100B, a large biomedical multimodal model with hundreds of billions of parameters. It is reported that based on the general language model, GhatDD-FM100B also superimposes biomedical expertise enhancement, multimodal alignment, and instruction fine-tuning and RLHF three-layer design to ensure that it can "truly understand the pharmaceutical industry." In 2023, the model underwent C-Eval evaluation and scored more than 90 points in four evaluations, including physician qualifications and basic medicine.

In addition, in terms of related algorithm technology innovation, Shuimu Molecular has also developed the LangCell single-cell and text cross-modal large model, the molecule and text cross-modal large model MV-Mol, and the atomic-level protein representation learning model ESM-AA, etc.; in 2023, the company successively open-sourced the lightweight scientific research versions BioMedGPT-1.6B and BioMedGPT-10B, mainly for scientific researchers to learn and use.

At present, based on existing technologies such as the GhatDD-FM100B base, the company has launched product-level applications for the pharmaceutical industry: the conversational drug development assistant ChatDD (Chat Drug Discovery & Design), which explores new drug development models by integrating and understanding multimodal data and conducting interactive human-computer collaboration with experts.

ChatDD, a conversational drug development assistant (illustration)

In terms of specific applications, ChatDD currently focuses on three scenarios: drug project establishment, preclinical research, and clinical trial assistant. Taking the rapidly progressing drug project establishment scenario as an example, this is an important starting point for drug research and development and BD, but the writing of project establishment reports often requires a large amount of tedious information collection and organization, such as target market competition and patent layout. At the same time, since project establishment work is difficult to outsource, the difficulty of information collection has also increased.

In this regard, ChatDD's participation can improve project efficiency and quality to a certain extent. At present, the company's cooperation with Fosun Pharma is mainly centered on assisting project decision-making, focusing on scenarios such as automatic intelligence analysis and commercial value assessment. "The internal feedback from customers is good," Nie Zaiqing revealed.

It is also reported that in preclinical research scenarios, large models mainly solve the problem of discovering new targets and new treatment plans. For example, Shuimu Molecular has reached a cooperation with the innovative Chinese medicine company Bio-Jingfang to explore the relationship between diseases in the field of Chinese medicine and different targets and signal pathways.

Clinical trials are the most expensive and risky part of drug development, and are also a difficulty that traditional AI technology has never been able to overcome. In this regard, the data analysis capabilities of large models may play a role in increasing the probability of success of clinical trials, such as assisting in finding more suitable patients for enrollment. Nie Zaiqing admitted that this is "a very interesting scenario, and we are investing a lot of effort in doing this." The company has already started cooperation with relevant clinical experts and CRO companies, but it is still in a relatively early stage.

"From the feedback from the market, I feel that both traditional pharmaceutical companies and Biotech are interested in using big models to participate in drug development, and have a certain willingness to pay. Our order cooperation is more in line with expectations. Big models are destined to be the future development trend. I believe that the ChatGPT moment in the field of biomedicine will soon come." Nie Zaiqing said.

Investor’s view:

Yang Lei, founding partner and managing partner of Huashan Capital, said: Shuimu Molecular team spans the two major scientific research fields of biomedicine and artificial intelligence big models, has many top scientific research experts in the industry, and has recruited senior industry experts to join, and the talent pool is still expanding. The development direction of Shuimu Molecular's ChatDD conversational biomedical R&D assistant is in line with the needs of the biomedical industry in the next decade. In the future, ChatDD is expected to play a role in the pre-, mid- and post-stages of pharmaceutical manufacturing, assisting business intelligence and project establishment, preclinical drug discovery, clinical trials and other links, and has extremely strong product competitiveness.

Sun Qi, founding managing partner of Daotong Investment, said: "The ChatDD fourth-generation drug development paradigm breaks through the limitations of AIDD, CADD and TMDD, connects human expert knowledge with large model knowledge, redefines the model of drug development, and provides new possibilities for efficient and accurate drug development. We are full of confidence in the future development of Shuimu Molecular, and we also look forward to the Shuimu Molecular team continuing to apply advanced algorithms to the encoding and interpretation of biological modal data such as proteins, DNA, and single cells, and in the long run help humans further open the door to data-driven life science discoveries."

Xu Jingming, Chairman of iFlytek Ventures, said: iFlytek Ventures has always adhered to the concept of industrial ecological investment. In the investment in Shuimu Molecule, we have seen good synergy between the two parties in the field of big models. The ChatDD product of the Shuimu Molecule team is a multimodal vertical big model focusing on the field of pharmaceutical R&D assistance, developed on the basis of the general text big model. The pharmaceutical R&D process involves a large amount of professional research analysis, document writing and other work, which have the opportunity to be accelerated by professional big models. Based on its professionalism and AI technology capabilities in the pharmaceutical field, Shuimu Molecule hopes to be the first in the industry to achieve labor cost savings and R&D efficiency improvements in the pharmaceutical R&D field.

Zhang Yu, head of Qingzhi Capital and Qingzhi Incubator, said: "We have always been optimistic about the positioning of Shuimu Molecule and have been paying close attention to its development in the long term. Shuimu Molecule has gained advantages in relevant technological innovation, data accumulation, product research and development, and market development, and has established industry barriers in key dimensions."