Related Reading:

Wang Jianhua from the Chinese Academy of Agricultural Sciences: From Antimicrobial Peptides to Breeding Improvement, Synthetic Biology Empowers Agricultural Technology and Industrial Breakthroughs

Academician Chen Jian of the Chinese Academy of Engineering: Synthetic Biology Reconstructs the 66 Billion Cosmetic Ingredient Chain, Domestic Brands Welcome New Opportunities for Breakthrough

“CurrentAIIn the field of biomanufacturing, the more mature applications are the 'three key steps': Pathfinding (designing biosynthesis pathways), Enzyme Mining (novel enzyme discovery), and Enzyme Engineering (enzyme sequence optimization)., this isAIThe relatively mature application scenarios, if in the futureAI"Strong enough to even have the hope of directly creating an enzyme that does not exist in nature based on chemical reactions." Recently, Wang Yifei, chairman of Zelixir (Shanghai Zhiyu Biotechnology Co., Ltd.),CEODr. Wang Sheng said in an exclusive interview with VCBeat.

According to the introduction,ZelixirFounded in2021Year, always focused on the frontier field of synthetic biology, deeply integrating artificial intelligence (AI) technology, focusing on three core businesses: synthetic biology R&D, high-value bioproducts production, and personalized customization services.

Zelixir Relies onAIEmpowered intelligent design of biosynthetic pathways, enzyme engineering, and microbial metabolic engineering, among other cutting-edge technologies,Successfully developed multiple innovative bio-based raw materials, covering fields such as food health, beauty cosmetics, healthcare, biopharmaceuticals, new materials, and agriculture.

Dr. Wang Sheng is one of the pioneers who first introduced artificial intelligence, especially deep learning, into the field of protein structure prediction., whose early work (2014-2017Year) forAlphaFold（2024The Nobel Prize) series has laid an important theoretical foundation and technical framework.

Dr. Wang Sheng holds a bachelor's degree from the School of Life Sciences at Shanghai Jiao Tong University and a Ph.D. from the Institute of Theoretical Physics at the Chinese Academy of Sciences. He has conducted in-depth research at top institutions such as the University of Chicago. After returning to China, he served at Tencent.AI LabSenior Research Expert, Focused onAIDriven Computational Biology. Dr. Wang Sheng is committed to promotingAI for ScienceThe integration of industry, academia, and research, he led the development offastAF2The algorithm has significantly improved the efficiency of new enzyme discovery and biosynthesis.

In Wang Sheng's view, pathfinding, enzyme mining, and enzyme modification are currentlyAIRelatively mature application scenarios in the field of biomanufacturing.

"Pathfinding" means leveragingAITo design biosynthetic pathways, thereby obtaining the final product. It is reported that "pathfinding" belongs to the process optimization phase, and the production process is one of the key aspects of biomanufacturing, potentially even determining whether the product can be scaled up and mass-produced.

Wang Sheng stated that the route design for biosynthesis must rely onAI, because if only through traditional database retrieval, it is difficult to break out of the previously existing synthetic routes, and only if humans have discovered this route, it may be possible to find it.

"By contrast,AIWith strong design capabilities, trained on data and learned molecular features,A synthetic route can be designed in a fanciful manner based on a given starting point and endpoint. This route may have never appeared in the literature, but in the eyes of experts, it might actually be feasible.”

In addition to "pathfinding,"AIIt also plays a significant role in the systematic management of biomanufacturing production processes. "This can be achieved through pressure, temperature,pH"And real-time monitoring of the fermentation process and metabolite quantities using sensors such as dissolved oxygen is also a key direction that Zelixir is currently developing," Wang Sheng told VCBeat.

Wang Sheng further pointed out that, in the separation and purification process,AICan also play two major roles:

First, during small-scale and pilot trials, throughAICalculatedPotential Excellent Process Solutions, but ultimately, it still needs to be tried to confirm.

Secondly, in the actual production process,AIReal-time monitoring of production and providing intelligent decision-making. If a failure occurs during unattended operation, as long as the sensor configuration is sufficiently rich,AIThis can minimize the danger and loss in the first time. For instance, excessively high pressure or overheating may cause the reaction to become uncontrollable, so it is necessary to...AIAdditional conditions: In case of similar situations, shut down the machine directly.TheseAIThe capability requirements may not be as high as those for the modified enzyme, and its applications may be more mature than those for the modified enzyme.

AIThere are indeed cases in the application of process optimization, such asPow.BioPlatform PassAIEnhance analysis to reveal the sources of bioprocess variation and provide early warnings of potential mutations caused by changes in bioreactor conditions.

According toLux researchAnalysis, in the process optimization phase of biomanufacturing,AIThe ability to detect subtle correlations in large datasets, generalize across operational ranges, and adjust control strategies in real time can enhance the effectiveness of traditional tools. IfAIAdvances in biomanufacturing processes that go beyond pattern recognition to develop into holistic, forward-looking control systems capable of detecting and responding to challenges such as contamination or metabolism are becoming crucial for biomanufacturing processes.

“AI"Will likely become increasingly necessary for bioprocess optimization in the next two to three years."Lux researchPoint out.

Enzyme mining is based on specific reaction steps, from knownProtein Sequence DatabaseTo mine the corresponding enzymes, similar to an "intelligent search engine".

According to Wang Sheng, the current collection of known protein sequences by humans has exceeded one billion, and the method to achieve enzyme mining is by leveragingAI, to mine a limited number of protein sequences with potential specified functions from a vast protein sequence database, and rely onDNASynthesis and molecular cloning techniques were ultimately validated for catalytic synthesis reaction efficiency in biochemical experiments.

The enzyme modification is due to the unsatisfactory performance of existing or mined enzymes, or they fail to meet the demands of industrial production. For instance, the substrate conversion rate may not be high enough, the activity not strong enough, or the enzyme’s thermal stability, soluble expression level, and other attributes fall short of expectations. In more challenging scenarios, it may even be necessary to alter the enzyme's product or substrate selectivity.AIBiocomputing technology can improve enzymes based on target attributes.

"The specific method is throughAIAlgorithm Communication 'Sequence-Structure-The relationship of 'function', seeking key amino acids that affect enzyme functional properties, and performing directed mutations.Traditional directed evolution methods mainly utilize a large number of saturation mutations, which are still effective and widely adopted for enzyme engineering, but with low efficiency and long cycles, relying onAIWill significantly improve efficiency."Wang Sheng pointed out.

But enzyme mining and enzyme engineering are important for biomanufacturing as well asAI, all seem to belong to relatively basic applications.

In Wang Sheng's view, the currentAIThe value of enzymes is similar to a "search engine," but their function is not fully utilized.The future must evolve from the "Search Era" to the "Enzyme Creation Era.", that is, given a chemical reaction, an enzyme capable of catalyzing it can be created, which would bring revolutionary changes to the industry.

"If it were just about casually creating an enzyme, many companies could do it, but if it’s about creating a catalytic enzyme for any given chemical reaction, only a handful of people globally can achieve that, and it’s basically impossible to apply industrially. So, we can only settle for the next best option, usingAI"Come to mine enzymes and modify enzymes." Wang Sheng said.

According to Wang Sheng, the difficulty of enzyme creation is extremely high. Current chemical synthesis technology has become very mature, almost at the point where "everything under the sun can be chemically synthesized." As long as the molecular formula of the substance to be synthesized is provided, chemistry experts can achieve synthesis through various methods. However, many reactions are either not efficient enough or cause significant pollution.

At this time, can we rely onAITargeted enzyme creation for greener and more efficient reactions?

Wang Sheng pointed out that the number of enzymes currently existing in nature that can achieve catalysis is not large. After millions or even billions of years of natural evolution, enzymes have become extremely elegant and efficient, but most of them can only function under normal temperature and pressure environments.

"In the future, humans may be able to directly design enzymes by combining a series of chemical reaction equations, which may not necessarily have to be20It is formed by non-standard amino acids, and it may not even be a protein in the traditional sense, but it can still catalyze reactions.But the challenge of enzyme creation is extremely high, with a difficulty level that could reach Nobel Prize standards."Wang Sheng pointed out.

To truly achieve enzyme creation, according to Wang Sheng's analysis, breakthroughs are still needed in the following four areas:

First, theoretical chemistry still requires some research, but not too much.

Second, the essence of enzyme catalysis is related to quantum mechanics, but current scientific exploration into the integration of quantum chemistry and biological catalysis remains insufficient;

Third, it is necessary to buildAIModels that can depict the most fundamental information of chemical reactions, involving dynamic reaction potential energy, electron transfer, bond formation and breaking processes, transition state characterization, etc., may only be described through quantum chemistry.But how to use these stepsAIThere is currently no algorithm that can achieve this in terms of language description and understanding.；

The fourth lies in verification, even preliminary.AIModel Birth: After designing hundreds of thousands of protein sequences based on a chemical reaction, only one sequence may catalyze the reaction.How to Conduct Efficient Validation, which is also where the difficulty lies.

AIThere are also broad application prospects in the field of cell factories.

A cell factory is a microorganism that, after being engineered from a chassis cell, can produce specific chemicals and proteins.

Wang Sheng stated that many synthetic biological products are produced through cell factories, but the improvement and creation of these cell factories still rely on traditional methods, namely continuous knocking out.DNADifferent fragments, or the continuous introduction of different exogenous sequences, or through a large number of random induced mutations, combined with high-throughput screening, although there are also many scientific bases,But in essence, it is more like relying on continuous trial and error to find the right solution.。

AIIt seems to improve the efficiency of "continuous trial and error," but there is currently no vertical large model for cell factories available.

According to Wang Sheng's analysis, the core of cellular factories lies in metabolic networks and pathways. In fact, humans have always wanted to make breakthroughs in cellular factories, but most efforts have only remained at the academic level, such as the highly popular virtual cells in recent years.virtual cell", essentially hoping to achieve throughAIModels to understand, depict, and predict cellular metabolic networks.

"Just likeAlpha Fold2Only after its groundbreaking emergence can humans grasp the relationship between the structure and function of sequential proteins.However, in the field of cell factories, there is still a lack of corresponding algorithms and models, or they are far from reaching a similar level.AlphaFold2The level makes it difficult to predict the metabolic network of cell factories."Wang Sheng said."

It is reported that, unlike enzymes, the metabolic network of cell factories is extremely complex, involving more than a dozen enzymes and simultaneously related toDNATranscription, reverse transcription, and metabolism form a systems engineering model. Moreover, the improvement goals of cellular factories often involve using the least materials, the shortest time, and achieving more output, which further increases the difficulty.

AndAIOnly by fully understanding this network can large models be built to truly empower cell factories.

Wang Sheng pointed out that a cellular factory is almost equivalent to an entire living organism, with its metabolic network encompassing not only the interaction network between proteins but also the interaction networks between proteins and nucleic acids, proteins and small molecules, and so on.

Moreover, metabolic networks change with external factors, and cells in small-scale,100Rise and100The metabolic state in tons of fermenters is completely different, which is related to cell density, environmental temperature,pHFactors such as values, stirring speed, and oxygen distribution are all related, and the difficulty in scaling up cell factories lies precisely in these issues.

"OnlyAIin order to understand and predict the metabolic network of cell factories, butAI"The empowerment in this area is still insufficient, and there is a lack of excellent modeling."Wang Sheng said.

Precisely because of this, Wang Sheng believesAIIn the application of enzymes, it will definitely be superior to cell engineering in the short term, and more controllable and efficient.

"Because the effect of the enzyme in a one-liter and a ten-thousand-liter fermenter may be basically the same, the enzyme we’ve excavated and modified shows a two-hour conversion rate in small reactions.99.8%, used in production10The same effect can be achieved in large-ton fermenters. Since enzymes originate from cells and are essentially catalysts, chemical catalysts rarely encounter difficulties in scaling up.

"But if we want to achieve a lower cost of biomanufacturing compared to chemical methods,Almost exclusively reliant on cellular factories,"But at present, the improvement of cell factories may still need to rely on traditional trial-and-error methods," said Wang Sheng.

Related Reading

"Ferulic Acid Vanillin" Successfully Achieves Industrialization from AI Computation to Hundred-Ton Scale

Lead the High-End Upgrade of the Food Flavor Industry

Recently, relying on its self-developed AI computing platform, Zelixir has successfully achieved a complete technical closed loop from molecular design to industrial implementation. Its first self-developed enzymatic natural fragrance product—ferulic vanillin—utilized an AI-driven enzymatic catalysis pathway, achieving stable mass production at the hundred-ton level within just one and a half years. The product has obtained EU Natural Certification and U.S. FEMA GRAS certification, qualifying it for sale in major global markets to meet the flavor enhancement demands of high-end and food industries. This advancement supports the transformation of the natural fragrance market towards green and high value-added directions.

Zelixir, with AI computing at its core, has built an innovative system of "synthetic biology + biocatalysis + green enzymatic processes," successfully replacing traditional chemical methods and plant extraction techniques. It has overcome several key biotransformation technologies, establishing a full-process industrial closed loop from raw materials—enzymatic catalysis—separation and purification—to stable mass production. This AI-powered biosynthesis pathway is internationally leading, creating a new model for natural fragrance manufacturing and breaking the long-standing technological barriers that relied on chemical routes and high-cost plant extraction.