De novo binder discovery
De novo protein binder discovery from target structure using computational design and experimental display screening
Start a binder project →AI protein binder design meets experimental validation
De novo binder discovery starts from a target structure — no immunization, no pre-existing library, no prior binder required. You provide a target. We generate, filter, synthesize, and screen binder candidates end-to-end.
Target analysis
We analyze your target structure to identify druggable surface patches, define hotspot residues for binding, and select constraints for the generative model. If no experimental structure exists, AlphaFold2 predictions are used.
Computational design
RFdiffusion, BindCraft, and Boltzgen generate diverse protein backbones in parallel, each suited to different scaffold types and binding geometries. ProteinMPNN and SolubleMPNN design sequences for each backbone. Thousands of candidates are generated and the best are selected across all campaigns.
Pooled synthesis and display
Top-ranked designs are synthesized as a pooled oligo library and cloned into yeast display vectors. The library is screened by FACS against fluorescently labeled target, with multiple rounds of increasing stringency.
Hit identification
NGS of enriched populations identifies confirmed binders. Enrichment ratios quantify relative binding strength. Top candidates are individually expressed, and binding is confirmed by flow cytometry titration.
Why de novo design for binder discovery
No immunization required
Traditional binder discovery requires immunizing animals or constructing large naive libraries. De novo design starts from the target structure alone. This is particularly valuable for conserved human targets that fail to elicit immune responses, or for toxic targets that cannot be used as immunogens.
Epitope-directed design
Computational design allows you to specify exactly which surface patch the binder should engage. This level of epitope control is difficult to achieve with immunization or library screening, where the immune system or random diversity determines which epitopes are targeted.
Non-antibody formats
De novo binders are not constrained to immunoglobulin folds. They can be designed as small, stable, single-domain proteins with tunable size, valency, and biophysical properties — advantages for intracellular delivery, diagnostic sensors, and non-clinical applications.
Speed and parallelism
Computational design generates thousands of candidate binders in days, not the months required for immunization campaigns. Multiple target sites can be explored simultaneously, and designs for different targets can run in parallel on GPU infrastructure.
Challenging targets
Flat protein surfaces, conserved viral epitopes, and protein-protein interaction interfaces that are difficult to target with traditional approaches are accessible to structure-conditioned generative models. De novo design excels precisely where conventional methods struggle.
Experimental validation included
Computational designs are hypotheses until experimentally confirmed. Our pipeline includes yeast display screening as an integral step, not an afterthought. You receive confirmed binders, not computational predictions.
Typical binder discovery outcomes
Outcomes depend on target difficulty, epitope accessibility, and the structural quality of the input model. Below are representative ranges from completed campaigns.
Computational candidates
1,000-10,000 backbone-sequence pairs generated per campaign. In silico filtering by predicted binding energy, structural metrics, and sequence quality reduces this to 100-500 candidates for synthesis.
Display-confirmed binders
Typical hit rates from pooled screening range from 5-50% of synthesized candidates showing detectable binding by FACS. Hit rate varies with target difficulty and the stringency of in silico filtering.
Affinity range
Initial de novo binders typically bind in the low micromolar to high nanomolar range. Further optimization by DMS-guided affinity maturation can improve Kd by 10-100 fold.
Sequence diversity
Multiple structurally distinct binder families targeting the same epitope. Sequence diversity among confirmed hits provides optionality for downstream development and IP positioning.
Deliverables
Ranked hit list with enrichment data, individual clone binding confirmation, sequence files, and structural models. Ready for downstream characterization, affinity maturation, or reformatting.
Timeline
6-8 weeks from target structure to confirmed binder sequences. Computational design: 1-2 weeks. Library synthesis and cloning: 2-3 weeks. Display screening and NGS: 2-3 weeks.
Have a target that needs a binder?
Share your target structure or PDB ID. We will assess feasibility and propose a de novo binder discovery campaign.
Start a project →