Signet Therapeutics and XtalPi: AI + Organoid Gastric Cancer Targeted Drug SIGX1094 Moves Toward Key Clinical Validation

“Five years ago, when we put this slide up, many people said we were just drawing a pie in the sky. But now the world’s first targeted drug for diffuse gastric cancer that we developed with XtalPi has moved into pivotal clinical validation.”

English text is being prepared.

English text is being prepared.

[Editor’s Note] The Prix Galien USA, known as the “Nobel Prize of the pharmaceutical world,” announced its 2026 award winners on October 29.

It is worth noting that innovative companies incubated by XtalPi have been shortlisted for this award for two consecutive years. Last year, Signet Therapeutics was nominated for “Best Biotechnology Product” for SIGX1094, the world’s first targeted drug for diffuse gastric cancer; this year, both Lemone and Moda were nominated for “Best Startup”—the former, with XtalPi’s R&D support, was nominated for a metabolism-enhanced CAR-T therapy at one-thousandth of the conventional dose, while the latter worked with XtalPi to develop MP-5342, the world’s first oral LDH inhibitor, targeting the global autoimmune market worth more than $100 billion.

In 2025, Signet Therapeutics, based on SIGX1094, the world’s first targeted drug for diffuse gastric cancer developed in collaboration with XtalPi, received a nomination for the Prix Galien “Best Biotechnology Product” award, becoming the only Chinese biopharmaceutical company nominated for that award that year, and also the world’s first AI-created molecule nominated for the Prix Galien.

The molecule number of this drug is 094, meaning that the “AI + organoid” R&D model had already found the most ideal candidate molecule by the time it synthesized the 94th molecule. This also means that the closed loop combining AI and robotics can find the target candidate molecule very quickly, within 100 molecules.

A year later, this FAK/SRC dual-target inhibitor for diffuse gastric cancer will complete Phase I clinical trials. Because it has received both orphan drug designation and fast track designation, it may be able to complete Phase II trials as early as next year and then be approved for marketing; the company’s Phase II/III combination therapy clinical trial (IND) with Innovent Biologics for KRAS-G12C mutation-positive non-small cell lung cancer (NSCLC) has officially been approved by the National Medical Products Administration (NMPA). The second pipeline, SIGX2649—a pan-TEAD inhibitor targeting the Hippo pathway—has received dual approval from Chinese and U.S. drug regulators and is expected to begin clinical trials in November. The clinical trials for the two pipelines are led by Professor Shen Lin, a leading authority in gastrointestinal oncology, and Professor Wu Yilong, a leading authority in lung cancer treatment, respectively.

What supports these advances is the “organoid + AI” technology platform jointly built by Signet and XtalPi. The BT+IT (biology + information technology) model they have formed together seeks to complete the closed loop of “AI molecule generation—automated synthesis—organoid testing—feedback optimization.” XtalPi is responsible for AI molecule generation and automated synthesis, while Signet uses organoids to complete disease model tests close to patients; the real efficacy and safety data obtained from experiments then flow back to the AI model, forming iterative structure-activity relationship optimization.

Zhuo Shu, CTO of Signet Therapeutics, believes that the core value of organoids is their ability to generate real data: by differentiating stem cells into tissues, the entire process data can be learned and organized by AI, bridging the gap between traditional animal models and humans.

Regulatory efforts are also advancing in parallel. The National Medical Products Administration is carrying out methodological validation of organoids as replacements for animal experiments. Signet, as the only company among the first batch of seven units, has completed validation work on five models: heart, kidney, liver, intestine, and tumor. The FDA is promoting the adoption of organoids and organ-on-a-chip, with its initiative driven by the U.S. Congress and required to report to Congress annually. In Zhuo Shu’s view, domestic regulatory progress is basically in sync with the FDA’s layout: “If we do not make the transition domestically, then when foreign pharmaceutical companies submit applications based on the new model, we will be unable to review them, and we will become decoupled from the international system.”

He recalled how Signet presented itself five years ago: “When we put this slide up back then, many people said we were drawing a pie in the sky and hyping concepts—now it is gradually working.”

The following is an excerpt from the conversation between Zhuo Shu, CTO of Signet Therapeutics, and Leiphone AI Technology Review, edited without changing the original meaning:

Zhuo Shu: The Phase I clinical trial of SIGX1094 is basically complete. We are also working with Innovent, using their KRAS-G12C inhibitor in combination with SIGX1094 to treat non-small cell lung cancer (NSCLC). The Phase II/III clinical trial (IND) has officially been approved by the National Medical Products Administration (NMPA). This is the second core indication SIGX1094 has expanded into after diffuse gastric cancer, marking the extension of SIGX1094’s clinical value from a single tumor type to a cross-tumor combination strategy based on pathway mechanisms.

In addition, in the United States, gastric cancer is a rare disease. We have obtained FDA orphan drug and fast track designations, which are expected to accelerate review and approval. In the United States, it may be possible to launch after completing Phase II clinical trials, saving about half a year. Because Phase III clinical trials inherently take a long time, and if there is no fast track, the launch process after completion often takes more than a year, with the entire cycle approaching three to five years.

Another pipeline, SIGX2649, has also just received dual IND approvals in China and the United States and is expected to begin clinical trials in November. The first pipeline’s clinical trial is led by Professor Shen Lin of Peking University Cancer Hospital, and the second pipeline is led by Professor Wu Yilong of Guangdong Provincial People’s Hospital. There are several more pipelines behind these, one in cardiovascular and others of different types, currently in preclinical stage. We will expand the organoid + AI pharmaceutical model from oncology to the fields of cardiovascular disease and kidney disease.

Zhuo Shu: The Phase II/III trial of the first pipeline and the Phase I trial of the second pipeline will both launch successively in November and December. For the first pipeline, we will also lay out clinical trials in the United States—gastric cancer is a rare disease in the US, and we have obtained both FDA orphan drug and fast track designations. With these, it is very likely that we can go to market in the US after completing Phase II clinical trials, and the cycle will be much shorter.

02

Zhuo Shu: Everyone knows the double-ten law of drug R&D; broken down, it comes from two reasons. First, manual work is too inefficient—the design space is narrow and the speed is slow. Second, there is the problem of biological models: in the past, drugs were tested using cells, nematodes, fruit flies, and mice, and there is a huge difference between them and the human body.

Returning to first principles: if you are making drugs for people, then use human things. Build a three-dimensional model in vitro that can simulate part of the structure and function of human tissues—this is the concept of organoids. The efficiency problem is solved with intelligence. This is the same as Musk building SpaceX: if every rocket launched is discarded, costs cannot come down; to reduce iteration costs, you have to recover them. From first principles, these are the two problems that must be solved for the pain points of pharmaceutical R&D.

Zhuo Shu: Five years ago, when we put this PPT slide out, many people said we were drawing a pie in the sky and hyping concepts: “AI hasn’t succeeded, automation hasn’t succeeded, organoids haven’t succeeded, and you put all three together?” They might have passed on us after one look. Now it is gradually running through.

This model naturally reached this point. XtalPi does AI and automation, and we do organoids. We don’t have those top-tier pharmaceutical experts, but with these new technologies, putting them together seemed logically feasible, so we might as well try. We also don’t carry a heavy burden—if it works, even better; if it fails, it is still an attempt. Looking back now, this model has indeed run through, and with high efficiency. The two pipelines we and XtalPi are developing through collaboration based on the “AI + organoids” model have successively moved toward the clinical stage, both progressing very quickly, and the data readouts are also very good.

Zhuo Shu: People feel there is a bottleneck, first, because there is not yet a drug that has been deeply involved in early discovery by AI or organoids and ultimately approved and successfully launched; second, because many people feel these tools are not as good in use as legend would have it. The threshold for pharmaceuticals is high, and the less familiar people are with this field, the more they feel the bubble is big.

We are inside the industry, step by step using these tools to push pipelines from 0 to the clinic: in many scenarios they are indeed not yet mature, but in mature scenarios—such as XtalPi’s crystal form prediction and enabling COVID drug development—the superiority of AI over the traditional manual experimental drug R&D model is already 100% certain. It is like new energy vehicles running 50 kilometers in the early days; it was impossible to go 1,000 kilometers as soon as they were invented. The same is true for AI and organoids: in many scenarios the data are not yet mature, which limits model training, but the direction is not wrong.

Zhuo Shu: Virtual cells and virtual organs are both directions worth exploring; the key is to find specific industrial application scenarios and continuously validate them in real problems. If the data do not sufficiently cover the target disease, cell state, or biological process, model predictions may deviate, and this is also where continuous improvement through experimental feedback is needed. XtalPi has also invested in companies in directions such as organoids, virtual cells, and organ-on-a-chip. Everyone has different emphases, and in close collaboration we apply these technologies to real scenarios in drug discovery, then continuously iterate based on experimental results and project feedback.

One of our core advantages is that we have our own ability to generate data; data on specific disease models, sustainably accumulated drug perturbations, and genetic perturbations are extremely scarce in public databases. These data can both support pipeline R&D and provide a basis for model training and validation.

For drugs that were successfully developed in the past, we understand their situations; but for candidate drugs that failed in R&D, the data are missing—we may know only that something went wrong at a certain step, but we do not know other properties or the entire biological effect. These failure data, extremely scarce in public literature, are also very important for drug R&D.

With organoids, we can differentiate a stem cell into a tissue and test a large number of failed and successful drugs, and the data from the entire process can be collected in large quantities; AI can then learn and organize these complex data well. This was impossible in the past—two-dimensional human cells could not even be cultured well in vitro, let alone three-dimensional tissue structures. These are all possibilities brought by organoid and AI technologies.

Zhuo Shu: After the target is determined, XtalPi performs AI-based molecule generation and design based on the target and carries out large-scale molecule screening around the target. For our first pipeline, the scaffolds screened involved hundreds of thousands of compounds, but very few actually needed synthetic validation, because virtual screening, activity prediction, and druggability prediction filter out all candidate molecules that do not meet the requirements.

In the past, the drug R&D model dominated by manual experiments had limited molecular design space, slow R&D cycles, and large gaps between animal models and humans. Now organoids are used for humanized testing and AI is used to improve efficiency—this is true dry-wet integration. Once this process is connected, it can replace the traditional R&D of many products related to biomedicine. Now it is indeed beginning to gradually generalize to more levels such as AI virtual cells, target prediction, and understanding of biological mechanisms. You can imagine that conventional AI target screening and biological mechanism understanding can now all be done.

Zhuo Shu: These are two very different technologies, and the cooperation structure determines whether the data can be connected. If you find an AI CRO company to cooperate with, it is unlikely that you will hand over all the core biological data to them; the costs of trust and communication are very high.

Our relationship with XtalPi is that XtalPi invests in and incubates us, holds a share of the pipeline revenue—there was a percentage in the early stage, and there is also a percentage after commercialization following listing. Our success is XtalPi's success; our interests are completely aligned, and there are no obstacles to the exchange of underlying data. If the cooperation remained merely a simple client-vendor model, I think this "long-termism" endeavor absolutely could not be pulled off. But this design of segmented interests and aligned stages will make both parties willing to go all in and get the hardest things done.

Now in the XtalPi ecosystem, our pipelines and those of Meda and Laimang are all advancing one by one, which has shown us the value of the collaborative R&D model.

04

Regulation and the Future

Zhuo Shu: Starting in 2024, we began working with the National Medical Products Administration to advance methodological validation of organoids and organ-on-a-chip. By the end of 2025, the first batch of validation results was completed, covering five models—heart, kidney, liver, intestine, and tumor—with a total of seven participating units, of which only we are a company; the others are research institutions such as Peking University, Tsinghua University, the Dalian Institute of Chemical Physics of the Chinese Academy of Sciences, the Institute of Zoology of the Chinese Academy of Sciences, and Southeast University.

It is precisely the support of the relatively good results from the first batch that led the National Medical Products Administration and the National Institutes for Food and Drug Control to successively establish innovation method evaluation centers this year and promote the implementation of the "Pioneer Plan," under which pharmaceutical companies can use new methodologies to evaluate drugs and submit reports to the NMPA.

Zhuo Shu: The FDA's work on organoids and organ-on-a-chip was initiated with project approval driven by Congress, and it must report progress to Congress every year, so its strategic priority is very high. China has joined the global review system. If foreign pharmaceutical companies apply for marketing based on the new model and their drugs are to enter China, domestic regulators must have the capacity to review them; otherwise they would become disconnected from the international system. The working meeting in May has already clarified the pace of progress, and an initial guideline will be implemented by early next year.

Zhuo Shu: Of course, the increase in visibility is very obvious, and attention from the capital market is also higher. Now state-owned capital will also come in directly, and the government is also paying close attention. The global chairman of the Galien Prize even recorded a special video at this year's Futian forum, explaining why Xige was nominated.

Zhuo Shu: I do not rule out this possibility. Once a pipeline reaches Phase II, it meets the conditions for listing, and we are also reviewing our internal structure.

Zhuo Shu: Right now we are still a pipeline biotech, but the greater potential in the future lies in our platform attributes. But I often use an analogy with investors: it is like Alipay and Taobao—at first it made money through Taobao, and Alipay was a tool; but once Alipay became a very strong financial tool, its value might be even greater than Taobao's. If our platform forms a very strong data infrastructure, its value may be greater than that of a single pipeline.

Zhuo Shu: From early 2022 to 2024 were the two hardest years for the industry. Valuations in the secondary market generally fell to one-tenth, and many biotech companies also went bankrupt. On the one hand, we increased revenue and cut expenditure as much as possible, but on the other hand, we also planned R&D funding in advance, so that we were fully prepared before the winter arrived and could prepare for long-term investment. The advantage of this innovation ecosystem is not only funding, but also strategic and business resources—if any one of these links goes wrong, many startups will face enormous risks.