Shuimu Molecular Selected for CCTV's "Top Ten Chinese AI Startups": The Only AI Drug Discovery Company, ChatDD Platform Already Used in Its Own Pipeline

CCTV's carefully selected "Top Ten Chinese AI Startups": the only AI pharmaceutical company is a Tsinghua-affiliated one. QbitAI Download the app for exclusive early access. November 02, 2026 12:30:00 From Beijing Updated November 02, 2026 17:01:13

Luyu, reporting from the Non-Conventional Hall

QbitAI | WeChat Official Account QbitAI

This time, it was CCTV itself that personally called the roll for the industry's next batch of star companies.

After 1,200 participating companies fought their way through round after round, the Top Ten Chinese AI Startups have freshly emerged.

Earlier, Moderna and Merck jointly announced that a personalized mRNA cancer vaccine developed with AI involvement had achieved positive results in the treatment of melanoma, long known as the "king of skin cancers."

Founded just over three years ago, the company was incubated by Tsinghua AIR, founded by Academician Zhang Ya-qin, with Professor Nie Zaiqing serving as chief scientist.

What Shuimu Molecular wants to solve is also the deep-water zone of drug R&D:

Since 2023, they have been continuously advancing the connection between natural language and biological language, and bringing the related capabilities into the conversational drug R&D platform ChatDD.

At the same time, it personally advances innovative drug pipelines targeting autoimmune diseases and tumors, applying the platform to its own R&D decisions, molecule design, and project operations.

Behind this multi-pronged approach is a plain judgment reflected by Shuimu Molecular and the entire pharmaceutical industry:

The ultimate R&D efficiency brought by this kind of human-machine collaboration must ultimately be cashed in as real clinical progress and success rates for China's innovative drugs.

Four generations of paradigm shifts in drug R&D

Shuimu Molecular's COO Xing Jie mentioned on the program that there is a widely circulated "three tens" rule in the pharmaceutical industry.

The birth of a new drug takes an average of 10 years and more than 1 billion US dollars in investment, yet the success rate is less than 10%.

The challenges go far beyond finding a molecule that can bind to a target.

How an innovative drug distributes in the human body after entering it, whether it can achieve the expected effect, whether there are unacceptable side effects, and which patients may benefit... Questions like these all require the artificial construction of a complete chain of evidence.

Models can predict structures, generate molecules, and then rank candidate results, but researchers may not fully understand the basis for the ranking, and it is also hard for them to feed their own concerns back to the model. Computational results can be handed over to experiments for validation, but between experts and algorithms, there is a lack of a channel for continuous collaboration.

Shuimu Molecular therefore proposed a four-stage division of drug R&D:

TMDD→CADD→AIDD→ChatDD.

TMDD (traditional methods of drug design) is easy to understand; it represents the earliest drug R&D process relying on accumulated experimental experience and expert judgment.

Later, computers began to participate in assistance, taking on part of the computation and simulation work, namely CADD (computer-aided drug design).

Further on, AIDD (AI-aided drug design) began to use AI for tasks such as property prediction and molecule generation. However, experts' experience and intuition were difficult to effectively interact with and organically integrate with AI models in the era of small models.

As for ChatDD (Chat-based drug design), what it tries to advance is enabling experts to participate in task planning, execution, and result correction through natural language dialogue.

In fact, this is also a rather forward-looking generational division.

ChatDD was proposed by Shuimu Molecular in 2023. Nowadays, overseas efforts such as GPT-Rosalind and Claude Science have also begun to explore interaction directions similar to it.

The scope of industry competition is evolving from single-model prediction capability to the organization of the entire scientific research workflow.

To put it in more accessible terms, it is about seeing who can become the entry point for the AgentOS of drug R&D first.

Dual-wheel drive of platform and pipeline

Shuimu Molecular is trying to feed these experiences separately into models and specific tasks, and let them continuously influence what to do next.

Concretely, it is an interaction system that continuously extends upward from the model foundation.

Letting paper knowledge participate in the next design

Take a simple example: suppose a researcher hopes to optimize a certain antibody through mutation. What the expert states is the functional goal, but what the model ultimately has to handle is the amino acid sequence and spatial structure.

There is clearly a language gap in between, and the first hurdle is the need to integrate the relevant knowledge into the foundation model.

BioMedGPT, which Shuimu Molecular advanced jointly with Tsinghua AIR in 2023, first explores the alignment of natural language with biological data such as molecules and proteins.

Researchers can ask questions around the input biological data, allowing the knowledge in the text to establish connections with specific research objects.

Among them, BioMedGPT-10B is also the world's first open-source, commercially usable 10-billion-parameter multimodal biomedical large model, ranking in the industry's first tier on benchmarks.

MutaPLM makes this connection more concrete.

This study, selected for NeurIPS 2024, maps textual descriptions of mutation effects to protein mutations, exploring both what changes a mutation may bring and proposing candidate mutations based on the goals described by researchers.

In short, the former is about understanding "if this position is changed, what may happen," while the latter asks, "if we want such a change to happen, where might we consider changing."

Mutation optimization also needs to consider the structural hurdle. The recently released StructEvo introduces structural information into protein mutation optimization, allowing candidate search to take spatial constraints into account.

According to the team, this research has been accepted by NeurIPS 2026.

Behind all of the above is a clear technical judgment from Shuimu Molecular: protein design has long faced a "trade-off" contradiction.

Changing a few amino acids may enhance binding ability, but it may interfere with the correct folding of the protein; increasing structural rigidity may improve stability, but it will sacrifice functional adaptability.

Therefore, around specific design tasks, the team developed corresponding models respectively. For example, CodeFP, accepted by ICML 2026, specifically focuses on the coordinated design of sequence and structure for functional proteins.

AbGaze further explores end-to-end de novo design of antibodies at the all-atom level, and uses this geometric information for sequence and structure generation.

...

Thus, from extracting knowledge from literature to translating functional goals into the basis for model analysis and design, the research results accumulated by experts over many years now have an entry point to participate in the next round of R&D.

The resulting AgentKG knowledge graph is used to organize domain knowledge and its associations, making every answer traceable and sourced.

Experts supplement goals, constraints, empirical knowledge, and questions through prompts; the agent plans tasks and calls tools accordingly, then returns the results.

After receiving the results, experts continue to ask follow-up questions or adjust conditions, and the next round of tasks changes accordingly. Then, through round after round of specific exchanges, experience begins to participate deeply in the computational process.

Turning single guidance into reusable methods

This issue pushes Shuimu Molecular's work from models and dialogue to skill packages.

Because expert experience is also reflected in research methods, such as what evidence to check first, when to supplement analysis, how to set screening conditions, and which steps must be reviewed by humans.

These processes, which can be clearly expressed and verified, also have the opportunity to be organized into reusable skill packages.

The first batch of OpenBioMed skill packages, just released this year, already includes 45 professional skills, covering five major categories: biochemistry/drug discovery, protein analysis and design, single-cell omics analysis, data retrieval, and tool invocation.

Among them, Skill Creator pushes expert participation into the workflow creation stage.

Researchers explain task intent, inputs and outputs, and constraints through dialogue; the system assists in organizing the process, then uses actual data to verify and modify it, forming skills suitable for the corresponding task.

In this way, a single guidance session from an expert has the opportunity to leave behind a continuously usable method.

Letting real tasks test the effectiveness of collaboration

Ultimately, this approach must also be tested in real tasks.

At the application layer, this means ChatDD for enterprises and Shuimu Zhiliao for individual researchers.

These correspond to two product models respectively: the former is top-down, where enterprises purchase and then promote employee use; the latter is bottom-up, more suitable for researchers and small teams.

First, ChatDD. Shuimu Molecular CEO Wang Zhuozhi describes its three main modules as Insight, Discovery, and Trial.

Insight first helps experts answer questions before project initiation: Is a target worth pursuing? Which patients can the resulting drug treat?

This requires simultaneously understanding disease mechanisms, unmet clinical needs, existing research evidence, competing projects, and patent status. ChatDD splits the related work among multiple specialized agents, collects and organizes information, and assists experts in forming judgments.

At Discovery, the task becomes how to design and virtually screen molecules.

Experts propose design goals, the platform organizes computation and candidate screening, and the results then enter researchers' judgment and experimental arrangements. This is also where Dr. Wang Zhuozhi believes revolutionary changes are emerging.

Trial extends the scope of support to clinical-related work, with the aim of improving the efficiency and success rate of clinical development.

Simply put, the most important significance of ChatDD is that, through dialogue, it enables life scientists to fully use AI tools.

In the past, life scientists often needed to learn programming or communicate repeatedly with computational teams to complete these operations. Now everyone can have an AI scientist team that truly understands drugs. Communication costs are reduced, and the efficiency of innovative drug R&D can be greatly improved.

As a biomedical expert, she chose the latter, namely Shuimu Molecular. Her judgment is:

In addition to project initiation and candidate drug design, ChatDD can also play a sufficiently large efficiency-improving role in downstream preclinical and clinical development stages, which are precisely the high-cost and time-consuming links.

Shuimu Molecular has already applied this capability to its own R&D projects.

Take Shuimu Molecular's first self-developed pipeline as an example. The project directly targets bispecific antibodies in the autoimmune field.

According to Pan Qi, when the project initially selected the target and R&D direction, it used ChatDD-Insight to analyze the disease field and competitive landscape, as well as preclinical, clinical, and translational research evidence.

After determining the direction, molecular discovery and optimization also continued to use ChatDD-Discovery.

From project initiation based on target biology and clinical evidence, to molecular design and virtual screening, to wet-lab validation and confirmation of the preclinical candidate molecule (PCC), this project took only seven months.

Compared with the 12-24 months required by traditional methods to obtain a preclinical candidate drug, ChatDD has already demonstrated the effectiveness of AI integration in early-stage R&D.

At the same time, project data also provides more direct feedback to the ChatDD platform, such as whether computational suggestions can hold up in experiments and how to adjust where they do not.

These questions will continually come back to the same R&D team, and the experts' experience, through this cycle of judgment, execution, and feedback, feeds into the next round of design.

This is precisely the value of the dual-engine drive of platform and pipeline:

Self-developed projects allow drug R&D experts to use the platform continuously, and also give the algorithm team the opportunity to track the downstream results of a task over the long term, continuously improving tools and processes.

Shuimu Molecular is gradually making this real across four dimensions: AI models, the agent platform and Skills, products, and pipelines.

While strengthening its foundations internally, it has not neglected the external side either.

Today ChatDD has also taken root with a number of external clients, for example the following types of collaboration:

Large domestic pharmaceutical companies: taking Fosun PharmAID as an example, Shuimu Molecular participates in co-creating the R&D agent platform, embedding AI capabilities into the enterprise's workflows;

Biotech companies: the platform serves leaner R&D teams, helping to build end-to-end agent-native innovative drug R&D processes, so that members can devote more energy to assessing direction;

CROs: collaboration extends further into the transformation of CMC operations, clinical trial protocols, clinical study reports, and regulatory submissions and related processes, empowering the CRO ecosystem with AI and positioning itself in pharmaceutical industry services;

Leading multinational pharmaceutical companies: strategic collaboration involves clinical operations, clinical submissions, post-marketing clinical work, and more.

Commercialization and open-source philosophy coexist

Seen this way, Shuimu Molecular is actually more like China's Isomorphic Labs.

The similarity is also reflected in their respective technical backgrounds:

But in contrast, Shuimu Molecular leans more toward large-molecule pipelines and open-source technology.

On the commercialization side, the twin growth curves of platform and pipeline advance in alternation, ensuring both the current pace of monetization and leaving room for a higher value ceiling.

2026 will be Shuimu Molecular's commercial scaling phase, focused mainly on standardized products with customized projects as a supplement. Xing Jie expects this year's revenue to be 3 to 4 times last year's, and the pipeline is expected to enter the development stage and business development next year.

Wang Zhuozhi previously served as vice president of WuXi Biologics, was deeply involved in the R&D of more than 300 innovative drug projects, and has work experience with Bayer's global collaboration; Pan Qi accumulated years of drug R&D and AI application experience at Genentech, Boehringer Ingelheim, Legend Biotech, and Pathos AI.

Xing Jie previously served as vice president and CIO of WuXi AppTec, and has long been involved in the digital and intelligent transformation of the pharmaceutical industry, investment and financing, industrial ecosystem operations, and commercialization.

This is also my deepest impression after talking with Shuimu Molecular: this team has both technical confidence and the dedication to turn ideals into reality.

Whether they are discussing models, products, or pipelines, their first principle is still the hope that technological progress will ultimately become good medicines that patients can use.

Starting from the scientific research accumulated at Tsinghua AIR and reaching the stage of CCTV's Ten Steeds, in three years Shuimu Molecular has turned its initial judgment into products and brought those products into real R&D.

But more worth looking forward to than embracing the next GPT moment of AI drug discovery is what kind of innovative drugs these changes will ultimately bring to patients.

The laboratory is the starting point; patients are where drug R&D must ultimately arrive.

"Special statement: The content above (including videos, pictures, or audio if any) is uploaded and published by users of Dafeng Hao, a self-media platform under Phoenix Net. This platform only provides information storage space services.

Notice: The content above (including the videos, pictures and audios if any) is uploaded and posted by the user of Dafeng Hao, which is a social media platform and merely provides information storage space services."