In the AI era, what 'innate challenges' does enzyme design actually aim to solve?

When it comes to enzymes, many people think of "biocatalysts"—mild, efficient, specific, and green. These advantages make enzyme catalysis a popular approach in fields such as pharmaceutical synthesis, green chemistry, and biomanufacturing. However, a fact that is often overlooked is: natural enzymes have evolved for the survival of organisms themselves, not for industrial production. When scientists try to bring enzymes into factories, they encounter four major challenges. In today's article, we will talk from the beginning—what exactly is enzyme design? What problems does it aim to solve? And where has it developed to now?

1. From "natural selection" to "on-demand intelligent creation"

The essence of enzyme design is the leap from "only being able to use natural enzymes" to "manufacturing artificial enzymes according to needs."

Natural enzymes are tools selected by nature over hundreds of millions of years, but their functions often do not align with industrial requirements. Enzyme design can be summarized in one sentence:

Change the amino acid sequence of the enzyme → change its three-dimensional structure → obtain new catalytic functions or better industrial performance.

Put simply: natural enzymes are tools with "factory settings," and enzyme design is about rewriting them into the shape you need—high-temperature resistance, resistance to organic solvents, catalyzing reactions that do not exist in nature, or increasing reaction speed by hundreds to thousands of times.

Professionally, enzyme design is usually divided into two directions:

1. Modifying known enzymes: On the basis of existing enzymes, make point mutations, fragment replacements, etc., to make them stronger (more stable, more efficient, with broader substrate range).

2. De novo design: Designing a completely new enzyme from scratch that can catalyze a specific reaction without relying on any natural template. This is equivalent to artificially creating a brand-new protein machine.

2. The 'Incompatibility' of Industrialization: The Four Major 'Tough Bones' in the Industry

If natural enzymes were already good enough, no one would go to such lengths. The reality of industrial demand is precisely what natural enzymes cannot achieve:

Poor stability: Industrial reactions often require high temperature, high salt, and organic solvents, while most natural enzymes lose activity within minutes above 60°C.

Insufficient selectivity: Natural enzymes may simultaneously catalyze several side reactions, resulting in impure products.

Insufficient activity: The catalytic efficiency toward the target substrate is too low; producing a single bottle of medicine may require tons of enzyme solution, making the cost unsustainable.

No corresponding enzyme activity exists: Some chemical transformations simply do not exist in nature, and no lead enzyme capable of catalyzing them can be found in natural enzyme libraries.

This is precisely the problem that 'de novo design' aims to solve. As early as 2008, David Baker's laboratory reported in *Nature* the successful design of 'designer enzymes' that catalyze reactions not yet found in nature. More recent progress is even more exciting: in 2023, they used deep learning to design a completely new luciferase, LuxSit, with activity, stability, and substrate specificity all superior to the natural version.

Pain Points of Enzyme Design

3. Real Applications of Enzyme Design: From Laboratory to Workshop

Once the aforementioned challenges are addressed, enzyme design can play a significant role in real industries. Here are several validated examples:

Drug Synthesis: For chiral sulfoxide drugs used to treat gastroesophageal reflux, traditional chemical methods rely on transition metal catalysis, which produces a lot of waste and is costly. The research team led by Yu Huilei at East China University of Science and Technology reprogrammed the catalytic pocket of Baeyer–Villiger monooxygenase, and the resulting mutant enzyme CbBVMOV3 achieved over 30 times higher catalytic efficiency. Compared to traditional chemical transition metal catalyzed synthesis processes, the enzyme-catalyzed process reduced waste emissions by 92.4%, lowered production costs by 80%, and increased space-time yield by 15 times.

Green Manufacturing of Fine Chemicals: The team led by Guo Ruiting at Hangzhou Normal University achieved "customized" halogenation of tryptophan through rational design of a natural single-component halogenase and developed a new "one-pot" strategy for synthesizing mixed halogenated tryptophan. This makes the green synthesis of complex halogenated drug molecules possible.

Synthesis of Non-Natural Amino Acids: The team led by Li Zhimin at East China University of Science and Technology constructed a multi-enzyme cascade system, starting from glycerol with water as the only by-product, achieving the conversion from inexpensive biomass to high-value functional molecules.

The common feature of these cases is that they all completed validation from the laboratory to industrial application. This was almost unimaginable five years ago.

Enzyme Engineering and Design

4. What is the biggest challenge in enzyme design?

Many newcomers to this field ask: Since there are only 20 amino acids and the sequence is fixed, why is designing an enzyme still so difficult? The difficulties are concentrated in three areas:

Sequence space is too vast

For an enzyme with 100 amino acids, the possible sequence combinations are 20¹⁰⁰, more than the number of atoms in the universe. You cannot try them one by one.

The mapping rule from 'sequence → structure → function' is incomplete

Sometimes we can predict whether a mutation will disrupt the structure, but it is very difficult to accurately predict whether it will enhance catalytic activity. Activity involves many subtle factors, such as enzyme flexibility, substrate access channels, and transition state stabilization.

Epistasis

Two mutations that are beneficial individually may, when combined in the same enzyme, either add up, cancel each other out, or even worsen the outcome. Predicting the effects of combined mutations is a classic challenge.

Traditional methods to solve these problems involve 'directed evolution' — random mutation and high-throughput screening. It is very effective but slow, labor-intensive, and expensive: performed in repeated rounds, often taking months or even years.

Challenges of Enzyme Design

5. How AI is Changing the Game — And the Arrival of 'Conversational' Enzyme Design

In the past three to five years, there has been a fundamental change in the field of enzyme design: using AI technologies such as deep learning to tackle the aforementioned challenges.

The logic behind this is not complicated:

Although the sequence space is vast, AI can learn patterns from a massive amount of known protein sequences and structures, providing predictions of 'where mutations might occur.'

For higher-order effects, AI can learn the synergistic relationships between different mutations, reducing the number of combinations that need experimental validation from millions to dozens.

For de novo design, AI can first generate a stable protein scaffold and then reverse-engineer the amino acid sequence, after which rapid validation can follow.

The key point is that AI is not meant to replace wet experiments but to equip experiments with an 'intelligent targeting system.'

In the past, directed evolution was like 'a blind cat trying to catch a dead mouse'; now, it has become 'tracking a steed with a map.' Efficiency improvements are often on the order of 10 to 100 times, and the timeframe can be reduced from a year to several weeks.

To allow more teams without deep computational backgrounds to leverage this capability, some integrated platforms have already appeared in the industry. For example, Matwings Tech has launched the conversational protein R&D agent MatwingsVenus™ (Xiaowu™). Users only need to describe the task goal in natural language (such as 'make this enzyme resistant to 70°C high temperature'), and the system automatically handles the full process of data retrieval, model prediction, and design optimization, directly linking to automated experimental validation. This effectively turns the previously exclusive 'AI wet lab' closed-loop capability, which only large teams could afford, into infrastructure accessible to individual researchers.

Enzyme design is transitioning from a highly specialized niche discipline to foundational infrastructure for the green manufacturing sector. If you are interested in enzyme design cases for specific applications (such as drug synthesis or plastic degradation), feel free to contact MatwingsVenus.