In the rapidly evolving landscape of synthetic biology, enzymes serve as the primary catalytic engines driving the bio-based economy. However, traditional enzyme engineering, reliant on labor-intensive directed evolution (DE) and trial-and-error rational design, often struggles to keep pace with the demanding requirements of modern industrial and pharmaceutical applications. CD BioGlyco is proud to offer its enzyme artificial intelligence (AI) design service, a cutting-edge solution that transcends the limitations of conventional protein engineering.
By integrating high-dimensional biological data with advanced machine learning (ML) algorithms, we provide a predictive and generative framework for creating high-performance biocatalysts. Our service focuses on the intelligent design of enzymes with superior activity, thermal stability, and substrate specificity, tailored specifically for the complexities of glycobiology and fermentation systems. At CD BioGlyco, we use AI to navigate the nearly infinite sequence space of proteins, identifying optimal variants with surgical precision and reducing the time-to-market for your bio-innovations.
Key Technologies
Generative Diffusion Models for De Novo Design
We utilize state-of-the-art diffusion-based generative models to design protein scaffolds from scratch. Unlike traditional methods that modify existing templates, these models build entire protein structures around a predefined active site (catalytic motif scaffolding), allowing for the creation of "synzymes" that catalyze non-natural reactions with high efficiency.
Protein Language Models (PLMs) and Transformers
Leveraging models trained on billions of evolutionary sequences, we employ PLMs to capture the "grammar" of protein folding and function. These transformers enable us to predict the functional impact of mutations without requiring extensive structural data, facilitating the discovery of distal mutations that enhance enzyme fitness.
Physics-Informed Deep Learning
To ensure the stability and expressibility of designed enzymes, we integrate geometric deep learning with molecular dynamics (MD) simulations. This "physics-aware" AI approach ensures that the predicted variants are not only theoretically active but also structurally robust and soluble under rigorous industrial conditions.
Enzyme AI Design: Bridging the Gap Between Digital Prediction and Biological Reality
At CD BioGlyco, our enzyme AI design service is a cornerstone of our chassis development and synthetic biology-based fermentation service. We recognize that the "bottleneck" in fermentation enzyme development is often the enzyme itself, its inability to withstand low pH, its inhibition by high product concentrations, or its low turnover rate. Our service scope is designed to address these specific hurdles through a data-driven, iterative process. We provide solutions for the development of enzymes involved in carbohydrate metabolism, glycosylation, and metabolic pathway flux. Our scope includes:
Enzyme Function Annotation and Discovery: We utilize contrastive learning and hierarchical multitask learning frameworks to annotate uncharacterized sequences from metagenomic databases. This allows us to "mine" nature for unique glycosyltransferases or hydrolases that possess the exact properties needed for your chassis.
Rational Stability Engineering: For enzymes used in high-temperature fermentation or harsh chemical environments, our AI tools predict mutations that optimize the folding free energy, enhancing thermostability and organic solvent tolerance.
Active Site and Substrate Specificity Optimization: Through AI-driven substrate-docking simulations and active site remodeling, we expand the substrate range of an enzyme or sharpen its regioselectivity, which is crucial for the synthesis of complex human milk oligosaccharides (HMOs) or glycoconjugates.
AI-Guided Directed Evolution: Even when laboratory-based evolution is necessary, CD BioGlyco uses ML to "rank" mutant libraries. Instead of screening 106 variants, our AI identifies the top 102 most promising candidates, reducing the experimental workload by 99% while increasing the hit rate for high-activity variants.
Workflow
Project Definition and Target Profiling
Every project begins with a deep dive into the client's specific requirements. We define the target catalytic parameters, including the desired turnover number, Michaelis constant, and environmental constraints such as optimal pH and temperature ranges.
Multi-Omics Data Aggregation
CD BioGlyco collects and pre-processes diverse datasets, including protein sequences, known crystal structures, and existing kinetic data. We leverage proprietary and public databases to build a robust training set specific to the enzyme class in question.
In Silico Library Generation and Design
Using our generative AI models, we design a virtual library of enzyme variants. This stage involves deep mutational scanning (DMS) predictions and scaffold optimization to ensure that the proposed sequences occupy a functional region of the protein fitness landscape.
High-Fidelity Performance Prediction
Each variant in the virtual library is filtered through multiple AI checkpoints. We evaluate structural stability, ligand binding affinity, and transition state stabilization using geometric graph neural networks to prioritize sequences with the highest probability of success.
Automated Build and High-Throughput Screening
Selected sequences are synthesized and expressed in optimized microbial hosts. CD BioGlyco utilizes automated liquid handling and microfluidic droplet-based screening (up to 107 variants) to experimentally validate the AI's predictions.
Iterative Feedback and Model Refinement
The experimental results are fed back into the AI platform. This "closed-loop" system allows the model to learn from "failed" variants as much as successful ones, refining its predictive accuracy for subsequent rounds of design if necessary.
Publication Data
DoI: 10.3390/molecules31010045
Journal: Molecules
IF: 4.6
Published: 2025
Results: This paper summarizes AI-driven enzyme engineering, highlighting how machine learning, deep learning, and generative models overcome the limitations of traditional methods. It introduces core techniques—including AlphaFold2, ESM-2, ProteinGAN, and reinforcement learning—for structure prediction, variant design, and de novo enzyme creation. The review covers advances in improving catalytic efficiency, substrate specificity, stability, solubility, and novel functions, alongside applications in pharmaceuticals, biofuels, food, detergents, and bioremediation. It also addresses key challenges such as data quality, model interpretability, and experimental validation, and outlines future directions toward standardized datasets, explainable AI, and automated hybrid workflows to accelerate next-generation biocatalyst development.
Fig.1 Overview of AI in enzyme engineering. (Khan, et al., 2025)
Applications
Pharmaceutical Glycoengineering
AI-designed glycosyltransferases are used to modify the glycan profiles of therapeutic antibodies, enhancing their effector functions and half-life in the human body for better outcomes.
Industrial Fermentation
We develop high-efficiency enzymes for the production of bio-based chemicals and fuels, optimizing the enzymes to function within the metabolic pathways of engineered microbial chassis.
Nutraceutical Production
Our AI design services enable the cost-effective synthesis of rare sugars and HMOs, ensuring high purity and yield for infant formula and dietary supplements.
Environmental Bioremediation
We engineer robust enzymes capable of breaking down complex environmental pollutants, such as plastic waste or agricultural runoff, under varied and challenging field conditions.
Advantages
Unprecedented Design Speed
Our AI-driven approach compresses years of traditional directed evolution into months, allowing for the rapid identification of lead enzyme candidates with optimized properties.
Expanded Sequence Diversity
By utilizing generative models, we explore "dark" sequence space that natural evolution hasn't reached, discovering novel functional motifs that traditional homology-based design would miss.
High Experimental Success Rate
Our physics-informed AI filtering ensures that the majority of synthesized variants are soluble and functional, drastically reducing the costs associated with failed laboratory experiments.
Precision at the Angstrom Level
We achieve high-resolution control over active site geometry, enabling the design of enzymes with exquisite regioselectivity and stereoselectivity for complex carbohydrate synthesis.
Frequently Asked Questions
Customer Review
"The team at CD BioGlyco revolutionized our HMO production pipeline. Their AI platform identified a glycosyltransferase variant with a 5-fold increase in turnover number that we had missed through three rounds of manual DE. The precision was remarkable."
– Dr. S.T., Principal Scientist.
"Working with CD BioGlyco's AI design service saved us nearly a year of R&D. We needed a thermostable amylase for a high-heat industrial process, and the top candidate from their first virtual library exceeded our stability requirements."
– W.T., Director of Bioengineering.
"What impressed me most was the integration. It wasn't just a list of sequences; they handled the synthesis and the HTP screening, delivering a fully validated lead candidate. Their expertise in glyco-enzymes is unparalleled."
By combining the predictive power of AI with our deep heritage in glycobiology, CD BioGlyco empowers researchers and industrial partners to move beyond the limitations of natural evolution. Whether you are looking to create a de novo catalyst or optimize an existing fermentation enzyme, our platform delivers the precision, speed, and reliability needed to succeed in today's competitive bio-market. Please feel free to contact us to accelerate your enzyme discovery.
Reference
Khan, M.F.; Khan, M.T. AI-driven enzyme engineering: emerging models and next-generation biotechnological applications. Molecules. 2025, 31(1): 45. (Open Access)
For research use only. Not intended for any clinical use.
Fig.1 Overview of AI in enzyme engineering. (Khan, et al., 2025)
