Hong Liang of Tianwu Tech: Cost, Success Rate, and Intellectual Property — Three Critical Hurdles for AI in Industrial Biomanufacturing | China Biomanufacturing 100, #7

Click to read the series of articles:Top 100 People in China's Biomanufacturing

"Based onAILarge Model, the enzyme we designedPHValue as high as14(i.e.1Mole/In an extremely strong alkaline environment (such as sodium hydroxide solution), it can still function stably.This directly breaks through the cognitive boundaries of traditional experience, and it is basically impossible to achieve based on existing experience and thinking.”

"With the help ofAI"Large models can achieve extraordinary results through sheer power. Although the specific mechanisms are still difficult to fully explain, from a results-oriented perspective, we have not only succeeded but also achieved industrial-scale production."

Recently, Distinguished Professor of Shanghai Jiao Tong University,Matwings TechnologyChief Scientist Liang Hong said in an exclusive interview with VCBeat.

Liang Hong, Distinguished Professor of Shanghai Jiao Tong University and Chief Scientist of Matwings Technology

From Hongliang's perspective, the national level is vigorously promoting the development of the artificial intelligence industry. This trend is not merely a simple policy endorsement but a profound strategic choice. In this strategic layout, biomanufacturing, as a typical AI application field, is receiving significant attention.

"As a major manufacturing country, China is undergoing a paradigm shift from traditional chemical manufacturing to biomanufacturing, but faces many technical bottlenecks in the process. The intervention of artificial intelligence is timely, not only greatly improving efficiency and reducing costs, but also achieving process innovations that are difficult for traditional technologies to breakthrough."

"This is precisely the core reason why the integration of artificial intelligence and biomanufacturing is currently receiving significant attention—on one hand, biomanufacturing itself is driving the transformation of the manufacturing system from reliance on land resources and chemical equipment towards a sustainable model; on the other hand, artificial intelligence, as a key technical means, is effectively solving core challenges within the biomanufacturing technology system, jointly propelling the manufacturing industry towards a green and efficient evolution."But at the same time,AIIn the practical application of the biomanufacturing field, it is necessary to overcome the three major challenges of cost, success rate, and intellectual property.”Hong Liang stated.

Matwings Technology was founded in2021Years, Focused onAIProtein design, dedicated to using general artificial intelligence technology, provides solutions for many fields such as biomedicine and synthetic biology.AIProtein Total Solution.2025Year8In the publication "Typical Application Cases of Artificial Intelligence in the Biomanufacturing Field (First Batch)" released by the Ministry of Industry and Information Technology in [Month], Matwings Technology's large protein engineering modelAIACCLBIO^®Successfully selected; 2025Year11During the CIIE in November, Bayer HealthCare signed a strategic cooperation agreement with Matwings Technology to continuously explore the application of intelligent protein molecular design and biomanufacturing innovations in the fields of gastrointestinal and skin health.

China, with its globally leading large-scale fermentation technology, has become an important force in the field of bio-manufacturing. Companies in China have already applied5Matwings Technology10Fermenters with a capacity of tens of thousands of liters, potentially ten times larger than similar international equipment, highlight China's significant production capacity advantages in the biomanufacturing sector.

However, in Hong Liang's view,The biomanufacturing industry still faces the challenge of being "large but not strong."Due to the industry's concentration mainly in the manufacturing segment, there is relative weakness in core product research and development as well as intellectual property layout. Particularly, in the design and modification of chassis components such as enzymes, this has led to limited profit margins for the industry.

In this regard, artificial intelligence technology is driving industrial transformation from two dimensions:On the one hand, by optimizing the production process to reduce costs and increase efficiency, on the other hand, by leveragingAIDesign and develop new enzymes or other compounds with independent intellectual property rights to help China's biomanufacturing transform and upgrade from a "production advantage" to an "innovation advantage."

VCBeat learned in interviews that many biomanufacturing companies are indeed optimistic about artificial intelligence technology but have doubts about its industrial application, thus tending to adopt a wait-and-see attitude. However, Hong Liang stated,AIThere have been many industrial application cases in the field of biomanufacturing.

"We are collaborating with an innovative pharmaceutical company in China, leveragingAIThe capability of large models successfully broke through the key technology for enhancing the alkali resistance of nanobodies; the alkali resistance of nanobodies was increased fourfold, and the service life was extended from two months to over half a year. The project has been completed.2024Year5"Monthly industrialization implementation, saving partner companies over ten million yuan annually."

In addition to the mining and optimization of compounds such as enzymes,AITechnology has also shown great potential in the production optimization process.

Through machine learning, it is possible to achieve control over the temperature, feeding, and ratio of the fermenter.pHReal-time optimization and precise control of key process parameters such as values can effectively increase the final yield of the target product without changing the core microbial strains and raw materials. Since this approach relies on a mature industrial big data analysis paradigm, the technical threshold is relatively manageable. It has successfully migrated from traditional manufacturing to the biochemical industry and is therefore considered by the industry to have promising popularization prospects.AILanding scenarios.

Whether it is "directed evolution" or "enzyme mining",AITo function effectively, it requires massive-scale protein sequence data.

Hong Liang stated that Matwings Technology'sAIThe large model for protein design is pre-trained on the world's largest dataset, which contains over90Hundreds of millions of protein sequences, including both public databases and private data from various extreme environments: temperature range from below zero to130Degrees Celsius, pressure from1to thousands of atmospheres,pHValue from1.0To11, Salinity from0To60%. These are all real protein sequences that exist in nature, and their environmental adaptability has been naturally verified, andAICompared with the newly generated proteins, they have higher reliability.

"At present, based on this vast dataset of protein sequences, ourAILarge-scale protein design models can directly infer the functional properties of proteins from their sequences, and this predictive capability can directly guide protein design:Instead of blindly constructing thousands of variants, weAIGenerate a small number of key designs that are most likely to achieve optimal functionality, and subsequently only need100The key test on the left and right can complete effective calibration.”

According to Hong Liang, the core logic of this method lies in: protein structure is more like intermediate information, while function is the ultimate goal. Traditional approaches rely on the complex derivation of "sequence → structure → function," whereas they have constructed a predictive model that goes directly from "sequence" to "function," making it possible to...AIAble to quickly complete the preliminary screening, significantly reducing the number and scope of experimental validations.

From this perspective, in the wave of deep integration of artificial intelligence and biomanufacturing, a kind of“AIPredictive Guidance Design+A Small Number of Experimental VerificationsA new R&D paradigm centered on this is emerging, and the innovation of this paradigm lies in its potential to completely transform the functional definition and validation venues for new protein products.

Hong Liang stated that the traditional R&D pathway involves high-throughput screening in the laboratory, followed by testing the selected candidate molecules in a complex industrial environment. If the process fails, it must return to the laboratory to start over, making the process lengthy and the iteration costs high.

The new paradigm, however, skips this cycle: starting with research and development based onAILarge models conduct preliminary design based on the understanding of massive data, followed by directly placing a small number of candidate samples in real industrial environments, such as specific temperatures, pH levels, or bio-reactive solutions with complex compositions for functional testing."Just a few dozen precise scenario-based experiments can accurately locate and optimize the function of proteins. This greatly shortens the path from laboratory to industrialization."

AlthoughAIIt is expected to disrupt traditional R&D paradigms, and there are already cases proving its industrialization capabilities. However, Hong Liang believes that in the current biomanufacturing industry environment, there are still three major challenges that cannot be ignored, which is also a concern for many companies regarding...AIThe main reason for still adopting a wait-and-see attitude.

First, companies in the current biomanufacturing field, especially those in China, are generally facing the reality of low profit margins.

Many domestic biomanufacturing companies are accustomed to adopting cost-reduction strategies to participate in market competition. Meanwhile,AIUnder immense profit pressure, if a startup attempts to offset its high R&D costs by raising service fees, customers often find it hard to accept — after all, in the already low-margin field of bio-manufacturing, cost control is essential for survival. As a result, "cost-saving" becomes a shared choice for both parties but also limits, to a certain extent, the application and development of high-quality technical services in the bio-manufacturing sector.

Secondly,AIThe success rate of innovation at the source of biomanufacturing is not100%Some companies have had failed experiences, which has affected their confidence in subsequent attempts.

According to the analysis, manyAIThe relative advantage of enterprises does not stem from the absolute maturity of the technology itself, but rather benefits from large-scale trial and error continuously conducted through academic research and corporate collaboration. This process relies on constant attempts and improvements in proficiency to reduce error rates and increase the likelihood of success, essentially following a "practice makes perfect" approach.

"Of course, this phenomenon is not only present in the field of biomanufacturing, but also in many otherAIThe success rate of enterprise projects is often lower, which also leads to some enterprises failing to achieve the expected returns after investment, and they may even feel like they are 'paying an intelligence tax.' It must be acknowledged that even the most advanced technologies and methods currently cannot guarantee 100% success, which is precisely the common challenge faced by the entire industry at this stage."

Finally, and most crucially, for biomanufacturing enterprises, protecting the core secrets of their own products is vital, as it directly relates to the lifeline of the company's survival.

The ultimate core product of a biotech company is often embodied in specific biological (gene) sequences, which are not only the heart of research and development but also the source of its market competitiveness. More importantly, these sequences, together with their production platforms, form the most critical assets of the company. Once handed over, the company faces significant risks in terms of intellectual property rights. Therefore, it should not be simply regarded as the company being "not open enough," because requiring it to disclose its core sequences is tantamount to asking it to give up the very "foundation" it relies on for survival.

"Therefore, to promoteAI"The deep integration with biomanufacturing requires a full understanding and response to the needs of enterprises in terms of cost, effectiveness, and data security."Hong Liang emphasized.

Although there are still three major challenges to face, Hongliang believes that future scientific research and industrial innovation will surely buildIn "Massive Data+Artificial Intelligence"On the basis of,AINot only does it have an unparalleled advantage in statistics and discovering patterns from massive amounts of data, but it can also be combined with automated equipment to form a powerful productive force.

"Faced with millions of articles and patents, human researchers can no longer cope alone.AILarge models can summarize the knowledge and lessons of our predecessors, allowing us to stand on the shoulders of giants while avoiding the detours they have taken.We don't have to dwell too much onAI"Why a certain design can be made is less important than verifying its output through experiments. As long as the results are effective and reliable, practice reveals true knowledge."

Hong Liang told VCBeat that, facingAIThe "high-dimensional" capabilities demonstrated in the field of biomanufacturing that surpass traditional experience indicate that both the scientific research community and the industrial sector should "follow the trend" and actively embrace this historic opportunity."WhenAIThe designed protein is able to1Mole/"When it remains intact after being soaked in a high concentration of sodium hydroxide for five days, this result itself is stronger than any eloquence."