Trump launched the "Genesis Project." Where are the opportunities for China's AI4S industry?

The course of history is often subtly altered during technological revolutions. When the astrolabes and sailing ships of the Age of Discovery gave way to precise coordinates and powerful engines, humanity's understanding of the physical world was completely reshaped. Today, scientific exploration stands at a similar crossroads: AI for Science is leading a revolution in scientific paradigms.

How is this transformation taking place? What "impossible" tasks is it actually solving? More importantly, in the grand narrative of global technological development, what future does it portend? At the Linear Capital AGM, three "explorers" from the frontiers of materials, biomedicine, and molecular discovery — Dr. Zhou Zhen from Yinghuachenrui, Dr. Liu Xiaole from Xunbaihui, and Dr. Duan Chenru from Shenduan Yuanli — shared their real - world experiences and profound insights as they navigate the uncharted waters of science on the new vessel of AI.

Their observations converge on a central point: In the global race of AI4S, Chinese companies or Chinese - founded startups have immense potential. When facing the "impossible" problems in fields such as materials, life sciences, and chemistry, which are highly complex and involve multiple intertwined variables, they can transform the long - standing trial - and - error process, which used to rely on intuition and chance, into a predictable, designable, and engineerable rational exploration. From the perspective of investment institutions, this may also be one of the most disruptive and commercially valuable investment themes in the next decade. Below are some highlights from this in - depth conversation:

What fundamental changes has AI brought to your respective fields? What impossible tasks is it solving?

 Zhou Zhen, Yinghuachenrui:

Our field of work is in the highly complex system of polymer materials. Traditional R & D cycles are extremely long, the cost of trial - and - error is high, and the success rate is very low. Take the ubiquitous plastic water bottles we use every day. This material was developed in 1941, but for a long time, it was only used in applications like fibers and films. It wasn't until 30 years later, with the accidental discovery of some processing techniques and crystallization technologies, that someone developed PE bottles. Therefore, we believe that AI will bring significant changes to the polymer materials field at three levels:

First, with the invention and iteration of AI technologies such as deep neural networks represented by Transformer, we will use AI through data - driven methods and algorithm training to better understand the interaction mechanisms of polymer materials at different scales. We will develop materials based on the structure - activity relationship mechanism, including the reverse design of some application solutions.

Second, it is the combination of AI, quantum mechanics, and molecular simulation. We can now obtain a lot of reliable underlying data through computational methods while maintaining high precision. This data supplementation will greatly assist in our model development and iteration.

Third, AI technology can bring about significant changes in the combined use of multimodal spectroscopy techniques. By using our multimodal technology in material characterization, we can establish large databases of different detection spectroscopies. At the same time, through the combined use of these techniques, we can conduct a more comprehensive and direct analysis of the structure of materials at different scales simultaneously. This will ultimately greatly contribute to the development of large - scale models for polymer materials. These large - scale models can not only accurately predict the properties of new materials but also perform reverse design based on performance requirements.

 Liu Xiaole, Xunbaihui:

We use AI in biomedicine. I've been conducting cancer research at Harvard University for 20 years and have always believed that insights into biology can significantly reduce the reliance on data and computing power. Through years of research on immunity, we've found that B - cells beside cancer cells in patients are actually producing anti - cancer antibodies, but the quantity is insufficient, similar to a "war - torn area" when a tumor forms. We analyzed the sequencing data of tens of thousands of tumors, which contained hundreds of millions of antibodies, and used AI to calculate which proteins these antibodies were targeting. So, what exactly can AI bring to us? Generally, people say "more, faster, better, and cheaper." In our field, it should be "precise, fast, better, and cheaper."

"Precise" means being able to find a completely unknown new target. Through analyzing the data of tens of thousands of tumors and hundreds of millions of antibodies, AI helped us discover that many patients' tumors produced a small amount of antibodies targeting a particular target. At that time, no one had studied this target, and there were fewer than 10 research papers worldwide on it. We were initially confused, but with a spirit of experimentation, we extracted the antibody from patients' tumors and injected it into animal tumors. To our surprise, it inhibited tumor growth. We then conducted reverse research on the mechanism while bringing the drug into clinical trials.

What does "fast" mean? In most cases, we need to thoroughly understand the function of a gene before starting drug development. However, we had already obtained the drug and knew it was effective in animals. We brought it into clinical trials while conducting reverse research on the mechanism. So, from the day we discovered the gene to its entry into clinical trials, it only took us three years, and it took five years to complete Phase I clinical trials. In contrast, it took Amgen 20 years to develop its PD - 1 drug from the start of research to Phase I clinical trials, while we only took five years. Moreover, our drug is the world's first to use AI to find a target, design an antibody, and reach clinical trials.

The third is "better." Compared with the uncertainties in traditional drug - development processes, our approach is based on the fact that many patients have already produced some antibodies. Therefore, the production is very stable, and the antibodies that enter the human body are naturally occurring, showing excellent performance. A patient who had tried five different drugs in seven months without success saw the disappearance of liver metastases after five months of using our drug, and the tumor shrank by more than 30% after seven months.

Finally, AI also helps us save a lot of money. In the United States, conducting a clinical trial on a cancer patient usually costs about $250,000 - $300,000. However, the actual cost for the hospital, patients, doctors, and drug administration is less than $100,000. The remaining more than $100,000 is spent on CRO companies and various consultants. We then started using computational methods for optimization, reducing the cost per patient enrolled in the trial by more than 40%. Recently, we've also been conducting clinical trials in China, which are faster, better, and cheaper.

 Duan Chenru, Shenduan Yuanli:

Both Dr. Liu and Dr. Zhou mentioned that the design space for chemical molecules and materials is extremely vast. Take the problems solved by AI in the past as an example. For instance, in AlphaGo, the 19x19 Go board has 361 possible moves, which is already the most complex among board games. However, the estimated design space for small - molecule drugs is 10 to the power of 60, meaning there are 10 to the power of 60 possible "moves" on the "board," which is an insurmountable problem for traditional AI.

We believe that a significant change occurred around 2022. With the advent of generative AI, we no longer need to consider the entire "board" when making a move. Instead, we can focus on the more critical and relevant chemical space by observing the surrounding environment as we make each move, allowing us to explore step - by - step in the process of molecular and material discovery.

One real - world case we recently conducted internally was to explore the reaction network starting from known substrates, similar to playing a continuous game of Go. Through this exploration, we discovered an interesting intermediate. Originally, this intermediate could only be synthesized by catalyzing a system with expensive raw materials and harsh reaction conditions. Now, we can achieve the same result under milder conditions without adding a catalyst. I believe that discovering such new reaction mechanisms is a crucial means of promoting the implementation of new materials and is a direction we've long valued.

From your practical experiences, what are the most core barriers for AI4S companies?

 Duan Chenru, Shenduan Yuanli:

In the past, there was a lot of attention on "scientists starting businesses," but it was later found that the success rate of scientists directly starting businesses was very low. From the perspective of scientists, their mindset often leans towards "creating a big splash" from a scientific point of view. However, the real world is like RL (Reinforcement Learning). We need to understand the surrounding world and establish more connections with the surrounding industries to obtain rewards and gradually drive the development of the business, rather than always aiming for immediate success and having everyone in the world buy our products.

Our startup officially began in June 2024. At the beginning, our most important barrier was our algorithm. However, a lot has changed in the past year and a half. The most important barrier has shifted from the initial algorithm to "how to create our own model and data iteration methods." Whether it's through data obtained from collaborating with customers or by refining a data flywheel from this data, this is a barrier we achieved within the first year of starting the business.

In the past six months, we've conducted numerous iterations and experiments with various leading customers. Whenever we try our best but still can't perfectly solve a problem, we often gain a lot of know - how, which is extremely valuable.

In the long run, I believe the most important factor is the team. Since our field is very new, our employees are also very young. Although I graduated with a doctorate not long ago, I'm already the fourth - oldest in the company. This is a highly interdisciplinary field. Our technical staff need to have both a scientific background and an AI background. When developing products or expanding the business, they need to understand both the customers and the technology. People need to communicate in different ways to make things work and establish a viable business model during scientific exploration.

 Liu Xiaole, Xunbaihui:

When we first started our business, people wondered how a single technology could bring about a qualitative change compared to others. However, the drug - development process is very long. We need to know how each step is carried out to use AI to make each step "precise, fast, better, and cheaper."

Some people think that in AI - based biomedicine, we can just sell software. Actually, large pharmaceutical companies won't spend a lot of money on software. They might offer only tens of thousands or hundreds of thousands of dollars. Others suggest that we can help large pharmaceutical companies design a molecule, but they might only pay a few hundred thousand or a million dollars. But think about it, which drug has been truly developed by a large pharmaceutical company in the past 20 years? Never. They usually take over from small companies when the drug reaches Phase I or II clinical trials. Their in - house R & D teams have never developed any molecules.

In this situation, if an AI - based pharmaceutical company only sells software or pre - clinical molecules, it will be very difficult to obtain large - scale funding. However, once a drug reaches clinical trials and shows effectiveness in patients during Phase I or II, it can strike a deal with a pharmaceutical company, potentially worth hundreds of millions or even billions of dollars. Therefore, we hope to use AI to optimize every step from early discovery to clinical development, at least in the early stages of clinical development, to maximize the company's value. Offering a comprehensive solution that is closest to the customers has the greatest commercial value.

 Zhou Zhen, Yinghuachenrui:

Dr. Duan and Dr. Liu have mentioned many capabilities that AI4S companies should possess, whether at the data level or in terms of capability integration. I fully agree. Specifically in the field of polymer materials, I believe another crucial factor is the understanding of the entire industrial chain. In fact, the chain from materials to applications is very long. For example, the clothes we wear go through a long process from materials to fibers, spinning, weaving, dyeing, and finally into finished garments. Each step involves a lot of industry know - how, which customers may not share with us. If we want to disrupt the industry or introduce something new, we must consider these aspects.

Therefore, for our AI4S field, in addition to the interaction and integration of materials and AI, I think more emphasis should be placed on engineering scale - up and understanding the industry in various fields. Ultimately, to transform from a technology - platform company to a commercially successful one, it requires the matching of various capabilities, like putting together a jigsaw puzzle; only when all the pieces fit can we see the complete picture.

Based on your observations in the past six months, what makes customers willing to pay for AI4S research results?

 Zhou Zhen, Yinghuachenrui:

Our company uses AI to empower polymer materials, and in the early stage, we focused on the 3D printing and bio - based fiber industries. In the past six months, the product that has most impressed customers and made them willing to pay is polylactic acid fiber. The biggest problem with this fiber is its long - term mechanical degradation, which causes it to become unusable at a certain stage in the textile industry.

To address this issue, we developed a very interesting hydrolysis - and - aging - resistant fiber material in the past six months. Compared with traditional PLA, under accelerated aging conditions, its mechanical properties remain unchanged for one month, which is equivalent to being able to be stored for 2 - 5 years under natural conditions. We developed a completely different product to address the industry's pain points.

 Liu Xiaole, Xunbaihui:

Actually, for our drug - development company, our direct customers are not cancer patients but large pharmaceutical companies. What large pharmaceutical companies find amazing is that we identified an unknown gene, quickly brought it into clinical trials, and it was effective in patients. They think this is very impressive.

We initially conducted all our Phase I clinical trials in the United States. In the past year, we've also started conducting clinical trials in China. We're not currently licensing our leading project to pharmaceutical companies because we expect to reach a value inflection point in the next year. Pharmaceutical companies then ask if we have other molecules. Since one molecule has shown promise, they wonder if there are more new ones. So, this year, we struck a deal with a pharmaceutical company to collaborate on some of our other molecules.

 Duan Chenru, Shenduan Yuanli:

We only started commercializing our products in the past six months. I'll share a case where we collaborated with a European and American beauty company to solve an "age - old problem," namely the stability of a certain active molecule in sunlight. This molecule was discovered in 1970, and its stability has never been resolved.

We conducted a POC project and recommended more than a dozen molecules to the company to stabilize this active ingredient. Initially, their R & D staff were very honest and said that if one of these dozen molecules worked, it would be amazing, and they'd give us an award. To our surprise, after laboratory tests, all the molecules we recommended outperformed their control group, and we subsequently initiated in - depth cooperation.

In the global AI4S race, what unique advantages does China have in this field?

 Liu Xiaole, Xunbaihui:

I believe that China has unique advantages in drug development: First, it's the talent in China. There are many people with extensive pharmaceutical experience in the United States, but there are a large number of newly graduated PhDs and post - docs in China. Moreover, the labor cost is lower than in the United States, and they work very hard and learn quickly.

Second, it's the overall infrastructure. For example, CRO companies that conduct reagent production, mouse experiments, toxicology, pharmacology, and clinical trials have a well - established foundation in China due to decades of industrial investment. When we tried to conduct some experiments in the United States, we found that it was more efficient to outsource the orders to China and have the results sent back. It's more efficient to conduct these experiments in Zhangjiang, Shanghai.

Third, it's the clinical resources