Ref K3
Computational Service
Confidence Reported
Scoped After Feasibility Review
Service Overview
AI-guided enzyme engineering replaces blind mutagenesis with focused libraries and model-driven rounds. We model the fitness landscape for your enzyme using sequence, structural and any measured variant data, predict which positions and substitutions are worth exploring for activity, selectivity or substrate scope, and design libraries small enough to screen properly. When assay data returns, models are retrained, which is what makes the second and third rounds considerably better than the first.
Key Features & Advantages
Libraries Designed, Not Randomised
Focused libraries at computationally selected positions cover the useful sequence space at a fraction of the screening burden of random mutagenesis.
Learning Across Rounds
Each round of measured data retrains the model, so the campaign converges rather than restarting; this is where machine-learning-guided evolution earns its cost.
Mechanism Kept In View
Catalytic geometry, transition state stabilisation and substrate access are modelled, so designs have a mechanistic rationale rather than resting on statistics alone.
Activity And Stability Balanced
Activity-improving mutations are frequently destabilising; both properties are predicted so the trade-off is managed deliberately.
Where This Service Is Used
1Improving catalytic activity or turnover for an industrial biocatalyst
2Changing enantioselectivity or regioselectivity of an enzymatic transformation
3Broadening or narrowing substrate scope toward a non-natural substrate
4Adapting an enzyme to process conditions such as temperature, pH or solvent
5Designing smart libraries to replace or supplement random directed evolution
Methods and Techniques Applied
1Structural and mechanistic analysis of the active site, substrate binding and access tunnels
2Substrate and transition state modelling, including docking and, where warranted, QM/MM treatment
3Position selection combining structural proximity, evolutionary coupling and predicted mutational tolerance
4Fitness landscape modelling with protein language models and supervised learning on available variant data
5Focused library design with codon and diversity specification for practical screening
6Round-by-round model retraining as measured screening data becomes available
What We Need From You
1Enzyme sequence and structure, with the reaction and substrate clearly specified
2Any existing variant data, however small, since even modest datasets substantially improve the models
3The improvement target: activity, selectivity, stability, substrate scope or process tolerance
4Screening throughput available, which determines how large the designed libraries should be
What You Receive
1Prioritised position list with the structural and evolutionary reasoning for each
2Designed variant sets or focused library specifications with codon schemes
3Structural models of the leading designs with substrate and transition state analysis
4Predicted activity and stability effects per variant
5A round-based campaign plan with data requirements for model retraining
What We Compute and Report
1Fitness predictions per variant with model validation statistics on held-out data
2Position prioritisation combining structural proximity, evolutionary coupling and tolerance
3Substrate and transition state poses in the active site, with tunnel and access analysis
4Predicted activity and stability effects reported separately so the trade-off is visible
5Library composition statistics: theoretical diversity, codon coverage and expected screening effort
Questions We Are Asked Before Scoping
How much variant data do we need?
▸None to start, since structure and evolutionary information alone support a first round. A few dozen measured variants improve the models noticeably, and several hundred allow genuinely predictive modelling.
Can computation replace directed evolution?
▸No, and it should not try. It reduces the size of the libraries that must be screened and increases hit density. Screening remains essential because activity prediction is still imprecise.
What improvement is realistic?
▸Order-of-magnitude improvements in activity or selectivity are achievable across a multi-round campaign for many systems, but outcomes are enzyme-specific and no responsible provider can promise a figure in advance.
Interpretation and Limitations
Computational predictions prioritise hypotheses; they do not independently establish binding, biological activity, therapeutic efficacy, safety or clinical validity. Performance depends on input quality, target class, the structural evidence available and the applicability domain of the selected methods. Scores used for ranking are comparative devices and should not be read as measured quantities. All deliverables are for research use only and are not intended for diagnostic or therapeutic use.
Is This the Right Service for You?
Choose this when
Activity, selectivity or substrate scope must improve beyond what random mutagenesis has achieved
Screening throughput is limited and libraries need to be small and smart
Assay data can be returned between rounds to retrain the models
Consider another service when
The enzyme is active enough but unstable
Protein Stability & Thermostability Engineering A completely new catalytic function is required
De Novo Protein Design What Goes Wrong, and How We Handle It
These are the failure modes that produce confident-looking results which do not survive the next experiment. Each one is designed against rather than discovered afterwards.
Activity predictions trusted at the same level as stability predictions
Activity prediction is substantially less accurate; computation is used to shrink libraries rather than to replace screening, and we say so.
Libraries designed beyond the screening capacity available
Library size is set from your measured throughput, so the designed diversity can actually be sampled.
Each round restarting rather than building on the last
Measured variant data is fed back and models retrained between rounds, which is where most of the gain comes from.
A Worked Example
Improving enantioselectivity in three rounds
The situation
A wild-type enzyme gave the wrong enantiomer preference and random mutagenesis had produced no useful variants.
What the work involved
The active site and substrate binding were modelled, positions were prioritised from structural proximity and evolutionary coupling, and a focused library was designed at a size the team could screen fully.
What came back
A library of roughly 1,500 variants instead of tens of thousands. Data from round one retrained the model for round two, and the campaign plan stated exactly what data each round required.
Illustrative of the shape this work takes. Details are generalised, no client or target is identified, and scope and outcome vary with the system in question.
What Published Work Reports
The results below were published by independent groups and are cited to show what these computational approaches have achieved in the literature. They are not ICDMO project results, and outcomes vary substantially between targets, datasets and programmes. We do not promise comparable results for any particular project.
Machine-learning-guided evolution raised a reaction yield from 12% to 93%
An active-learning workflow optimising five interacting active-site residues improved the yield of a desired non-native cyclopropanation product from 12% to 93% in three rounds of laboratory experimentation. The design principle is the one we apply: computation chooses which small set of variants to make, and measured results retrain the model between rounds.
Nature Communications, 2025 ↗1.6- to 42-fold activity improvements from model-selected enzyme variants
A machine-learning-guided enzyme engineering campaign evaluated 1,217 variants across 10,953 reactions to build predictive models for amide synthetases. Model-selected variants showed 1.6- to 42-fold improved activity relative to the parent enzyme across nine pharmaceutical targets, illustrating both the scale of data that makes models predictive and the spread of outcomes that remains.
Nature Communications, 2024 ↗De novo binder design succeeds at rates that vary enormously by target
Recent design methods report nanomolar binder hit rates between roughly 17% and 82% across evaluated targets, while independent evaluations of earlier pipelines on other targets report designs failing on expression, non-specific binding or undetectable affinity. The honest reading is that de novo binder design now works, that success is strongly target-dependent, and that panels must be sized accordingly rather than on a best-case figure.
Comparative design studies, bioRxiv, 2025 ↗Service Workflow
1
Scientific Consultation
We discuss the objective, the decision it supports and the data you hold. Nothing is quoted before we understand what the result has to answer.
2
Feasibility Review
We check that the data, target class and available structural evidence can support the proposed approach, and say plainly where they cannot.
3
Technical Proposal
Scope, methods, deliverables, timeline and price are confirmed in writing, together with the limitations we expect to report.
4
Computation and QC
The analysis runs on pinned tool versions in a containerised environment, with quality checkpoints and method comparison where that is informative.
5
Report and Follow-Up
You receive the ranked results, the documented methods, the limitations and a proposed validation plan, plus support in interpreting them.
Service Process
1
Online Consultation
›
2
Solution Matching
›
3
Service Contract
›
4
Project Execution
›
5
Data Delivery
Specifications & Turnaround
Note: ICDMO delivers computational (in silico) services. Results prioritise hypotheses for testing; they do not by themselves establish binding, biological activity, efficacy or safety. All deliverables are for research use only and are not intended for diagnostic or therapeutic use. Downstream experimental work is scoped separately and executed through qualified partners.
Get a Custom Quote
Our scientific team responds within 24 hours with a detailed technical proposal and pricing tailored to your project.
Contact Us Online ConsultationStandard Deliverables
✓Sequence-verified final product
✓Full QC documentation
✓Certificate of Analysis (CoA)
✓Electronic data files
✓Project completion report
Why ICDMO?
Computational CDMO Focus
Reproducible, Versioned Pipelines
Confidence & Limitations Reported
Experiment-Aware Recommendations
Response Within One Business Day