From target characterization to AI-designed protein binders
A computational path from difficult protein targets to validation-ready binder candidates
When naturally occurring binders are unavailable and structural information is limited, finding where and how to bind a protein can become a major discovery bottleneck.
Ardigen combined AI surface mapping, structural modeling and physics-based methods to identify new binding regions, generate more than 100,000 candidate scaffolds and progressively narrow them to a set of 100 high-confidence protein binders ready for in vitro validation.
The Challenge
Designing binders when obvious starting points are missing
The project began with three protein targets for which conventional binder discovery offered limited starting information.
The key challenges were:
- Lack of crystal structures, limiting direct structure-guided design.
- No naturally occurring protein binders available as starting templates.
- The need to identify previously unexplored binding regions on the target proteins.
- A very large potential design space that could not be tested experimentally candidate by candidate.
- The need to prioritize binders according to defined structural and biochemical properties before committing resources to wet-lab validation.
The decision was not simply which sequences can be generated? It was which binding regions and candidate binders provide the strongest basis for experimental validation?
Approach
From protein surface analysis to ranked binder candidates
Ardigen built an integrated computational workflow connecting target characterization, AI-based generation and physics-informed candidate selection.
1. Identify promising binding regions
AI-based surface mapping was used to characterize target surfaces and identify potential protein-binding regions.
This step identified two novel binding regions across the three studied proteins, creating new starting points for binder generation.
2. Generate candidate protein binders
Once potential interaction regions were defined, AI-driven de novo design was used to explore a much larger candidate space than would be practical experimentally.
The workflow generated more than 100,000 AI-designed scaffolds for further computational assessment.
3. Filter candidates across multiple properties
Rather than ranking candidates on a single predicted interaction score, Ardigen applied a cascade of computational filters including:
- sequence filtering
- pI and thermal stability assessment
- co-folding and structural ranking
- solubility filtering
- interaction quality ranking
The workflow also incorporated docking, de novo design, reverse folding and physics-based evaluation to progressively reduce the candidate space.
4. Prioritize candidates for experimental validation
The final output was a shortlist of 100 high-confidence binder candidates, selected according to the agreed biochemical and structural criteria and prepared for subsequent in vitro validation.
Results
More design space explored. Fewer candidates taken into the lab.
2 novel binding regions identified across the three studied proteins
100,000+ AI-designed scaffolds evaluated and narrowed to 100 high-confidence binders
3× faster than the classical wet-lab-based approach
Up to 1 month from target to computationally prioritized binder candidates
The outcome was a ranked set of physics-validated protein binder candidates ready for in vitro testing, reducing the experimental search space before wet-lab validation.
From computational design to a focused experimental shortlist
The value of computational binder generation is not simply producing more sequences. The workflow must progressively answer three questions:
Where can the target be bound?
AI-based surface mapping identifies plausible interaction regions.
What could bind there?
Generative models explore candidate structures and sequences beyond known natural binders.
Which candidates are worth testing?
Structural, physicochemical and interaction-based filters reduce a very large computational search space to a focused experimental set.
For this project, that meant moving from more than 100,000 computational designs to 100 candidates ready for experimental validation.
Have a protein target but no obvious binder starting point?
Further reading from Ardigen’s Knowledge Hub
- AI for Biologics
Explore Ardigen’s broader capabilities in target characterization, binder generation, hit screening and lead optimization. AI-Driven De Novo Generation of Protein Binders
Explore the scientific methodology behind AI-driven binder generation, including binding-site identification, protein design and structural validation.
Designing high-affinity peptide binders with AI
See how AI-supported de novo design can also be applied to generate and prioritize target-specific peptide candidates.
