Zhiyu BioTech Completes 18-Month Journey from AI-Driven Design to Industrial-Scale Production of Natural Flavor Ingredients

On November 30, 2020, the artificial intelligence system AlphaFold2 took the crown in the 14th Critical Assessment of Protein Structure Prediction (CASP) competition. Nature News even titled it “It will change everything.”

In 2024, the Nobel Prize in Chemistry was awarded to David Baker (University of Washington, USA), Demis Hassabis, and John Jumper (DeepMind) for their groundbreaking contributions to the field of protein structure prediction, particularly for revolutionizing protein structure research by combining artificial intelligence (AI) with computational chemistry.

AlphaFold2 has taken a significant step forward in the field of protein structure, often regarded as the "jewel in the crown" of computational biology, inspiring many peers—Upon the release of AlphaFold2's prediction results, Dr. Wang Sheng, then a senior research expert at Tencent AI Lab, realized that "a new era had arrived."

As a top scholar globally in the field of artificial intelligence combined with protein research, Dr. Wang Sheng has over a decade of expertise in protein structure prediction and computation, and was ranked first in the CASP12/14 protein contact map prediction. During his postdoctoral fellowship at the Toyota Technological Institute at the University of Chicago, Dr. Wang Sheng had frequent and close academic exchanges with John Jumper. He participated multiple times in collaborative discussions with John Jumper's research group and co-authored an article published in the Biophysical Journal.

In 2021, Dr. Wang Sheng and his team founded Zelixir (Shanghai Zhiyu Biotechnology Co., Ltd.), exploring the application of AI biocomputing in the field of synthetic biology. In 2023, Dr. Wu Dong's team, with extensive experience in precise product selection and having led a team to achieve full-line penetration from "experiment - pilot test - mass production," joined Zelixir, deeply integrating the advantages of AI biocomputing with synthetic biology product selection.

In June 2024, the enzymatic natural fragrance products independently developed obtained FEMA GRAS certification under the FDA system, which can be sold to most regions globally, meeting the demand for flavor upgrades in high-end and food industries. On December 5, the green manufacturing plant located in Changzhou, Jiangsu, and the hundred-ton-level bio-synthetic natural fragrance production center were completed. Zelixir officially entered a new phase of industrial application and production.

When AI protein design enters synthetic biology, how to find the product selection entry point? And how to make AI technology effectively land in R&D - small-scale trial - pilot-scale trial - industrial mass production? Taking this opportunity, VCBeat interviewed Dr. Wang Sheng to discuss how to create a new global paradigm for this biomanufacturing startup, Zelixir.

(Dr. Wang Sheng, Founder and CEO of Zelixir)

To describe the disruption brought by AlphaFold2 in one sentence, Dr. Wang Sheng said: "It has elevated protein structure prediction to the accuracy of wet lab experiments, with much higher throughput and efficiency than experimentation, implying ultra-low prediction costs and ushering in a new era for the application of protein prediction."

As a disruptive technology and tool, AI will boost the synthetic biology industry to phase 3.0, which involves intervention in the entire process of bio-manufacturing from computation (IT) → research and development (BT) → mass production (VT), namely Zelixir's self-developed "T³ Closed-loop" business model. In short, compared with the original R&D model based on repeated experiments, Zelixir's self-developed platform conducts R&D generation through AI at the computational end first, followed by experimental validation.

In terms of R&D computing, Zelixir has innovatively introduced an "on-demand enzyme design" pathway based on protein structure prediction theory. Specifically, this approach does not use traditional methods for enzyme design but directly modifies the key component of enzymes—proteins. The overall process is divided into "three key steps": "Pathfinding" – "Enzyme Mining" – "Enzyme Modification," thereby achieving optimized enzyme synthesis.

 (Zelixir Automation Laboratory)

On the one hand, this AI-based enzyme engineering R&D model can significantly reduce pollution in the chemical synthesis process, improve resource utilization efficiency, decrease environmental pollution, and achieve green, intelligent manufacturing. On the other hand, the industrialization cycle is notably shortened, enhancing product quality and safety while establishing an efficient and precise design pathway.

Led by a core team with dual expertise in both biology and AI, Zelixir successfully mapped out enzyme synthesis pathways in its early products and accumulated extensive data and algorithm iterations through collaborations in bio-computation and biopharmaceuticals. However, returning to the biomanufacturing sector, the challenge of "product selection difficulty" remains a persistent industry pain point for Zelixir. At the end of 2022, Dr. Wu Dong, who joined the company as a strategic advisor, brought new hope. With years of deep involvement in biomanufacturing, Dr. Wu Dong has rich market insights and product selection experience, along with a track record of continuous successful entrepreneurship, spanning sectors such as food additives, medical aesthetics raw materials, and health supplements.

"AI is very important, but at least for now, it is still a technology and a tool. The advantage of the AI team lies in their excellent mastery of the technology and their possession of a certain biological background, but AI cannot replace Dr. Wu's team’s keen insight into the market, implementation, and products. Only by combining the two can this path be successfully navigated," mentioned Dr. Wang Sheng. The first product co-developed with Dr. Wu Dong's team is a natural fragrance product that received FEMA GRAS certification under the FDA system in June 2024, which can be sold to most regions globally in the future.

It is reported that the product has effectively solved the long-standing problem of impurity odor in chemical synthesis methods. With its excellent performance and green, sustainable features, it provides the market with a healthier and more environmentally friendly choice.

Dr. Wang Sheng mentioned that this enzymatic natural spice product fully validated the efficiency of the "pathfinding → enzyme mining → enzyme modification" approach: laboratory route verification was completed in less than three months, and it quickly advanced to the small-scale trial and pilot-scale trial stages. Pilot trials began in November, and FDA certification was completed and passed by June this year.

 (Data source: Internal estimates by Zelixir, for reference only)

The key to advancing product experimentation at "super speed" lies in using AI to iteratively modify and optimize the various attributes of enzymes for application-specific needs. This requires close collaboration between the AI team and the experimental team.

In terms of R&D paradigm, Zelixir innovatively carries out process innovation, transforming the previous linear iteration into cross-parallel lines, allowing the AI team to deeply accompany and intervene in all aspects of the experiment."In a second-phase experiment, the reaction rate of the candidate enzyme mined by the AI team was extremely low. Although the lab pathway was successfully established, it definitely could not support small-scale and pilot trials. With the rapid intervention of the AI team, the reaction rate was increased from 5% conversion in 72 hours to 98% conversion in 2 hours." Wang Sheng mentioned that the traditional linear approach often implies disconnection. If problems with candidate products are only discovered during the scale-up or mass production stages, they are traced back layer by layer according to the experimental pathway, leading to repeated ineffective experiments and prolonged R&D cycles.

"At the pilot stage, more variables such as pH value, temperature, and organic solvent tolerance will affect the yield and productivity of the candidate enzymes, leading to evaluation issues. After the experimental data is obtained, the pilot team provides direct feedback to the AI team immediately. Through our self-developed algorithm and several rounds of iteration, we can achieve industrial-level improvements in thermostability, organic solvent tolerance, and solubility without affecting the conversion rate. Subsequently, the pilot team may pass the test in one go."

Furthermore, this November, at Fi Europe, the international event for food ingredients and functional foods, Zelixir's product received widespread recognition. It is the fastest and most efficient product route in the global development of such spice products using enzymatic innovation.Dr. Wang Sheng emphasized, "In the enzymatic route, Zelixir's products are at the forefront globally, having fully integrated AI-powered processes and moved into the final stage of product scaling and production, truly achieving a groundbreaking leap in AI-driven biomanufacturing from zero to one, placing them at the cutting edge worldwide."

(Zelixir Production Workshop)

Backed by its product and team strength, Zelixir is building a large comprehensive production base against market trends. Covering an area of approximately 4,500 square meters, the plant is expected to produce hundreds of tons annually with an annual output value of over hundreds of millions of yuan after official operation. As a comprehensive production base, it includes several core areas such as a pilot workshop and a hundred-ton-level biosynthetic natural fragrance production center, equipped with advanced production facilities and a top-notch technical team.

Dr. Wang Sheng talked about how the production base investment was made after repeated considerations and thorough preparations. First, it is based on the protection of breakthrough and disruptive technologies. Up to now, Zelixir has been granted eight invention patents, and 14 patent applications are in the acceptance stage. Secondly, during the small-scale trial and pilot-scale trial stages, Zelixir innovated the process route and carried out specialized modifications on some equipment. Therefore, corresponding “tailor-made” personalized solutions have also been implemented in the production base. Finally, the first natural fragrance product pipeline will be built as Zelixir's demonstration production line, showcasing a breakthrough from zero to one within a controllable cost range.

Since its establishment in 2021, Zelixir has achieved revenue in collaborations such as AI biocomputing empowering biomedicine, relying on its robust underlying algorithm platform and the experience of AI Lab, providing continuous and stable cash flow for R&D advancement.

"In the first two years, we need time to refine and polish our tools, and we also hope to find a suitable biomanufacturing implementation team. Therefore, we chose to collaborate with some well-known pharmaceutical public companies and Biotech in China on co-creation, using a set of basic modules for foundational predictive tasks such as protein structure prediction, small molecule structure, and protein co-folding. Relevant pipelines have already entered the research and development stage," said Dr. Wang Sheng.

As biomanufacturing enters a new phase of mass production, Zelixir will focus on bio-based materials as its core growth curve and explore a co-creation model that combines bio-computing with wet-lab experimental capabilities.

In 2023, Zelixir led the "The project "Creation, Application, and Information Sharing of P450 Catalytic Components Based on Big Data and Artificial Intelligence" has been successfully selected as a "National Key Project."Looking ahead to the industry front, Dr. Wang Sheng and Zelixir hope to do more: multiple startups with different strengths and experiences join hands to grow the pie of bio-manufacturing together. Meanwhile, across China, several governments are building pilot platforms and industrialization platforms to integrate resources, support each other, complement each other, achieve win-win results, and promote the progress of regulations and supervision.

In the future, Zelixir hopes to work together with biomanufacturing partners at all levels to further integrate AI into every aspect of biomanufacturing, such as enzyme catalysis, post-processing techniques, and quality inspection. This would achieve full-process digitization and informatization, ultimately leading to unmanned "dark factories" in biom manufacturing. "Ultimately, we believe that disruptive technologies not only drive progress in products and industries but, more importantly, achieve green sustainability, solving the industrial challenges of high pollution and high energy consumption."