Start from Scratch. Explore New Antibody Designs.
De novo design starts without a known binder: a target antigen, a research objective, and no antibody lead to improve. We generate candidate sequences, predict how each sits against the target, and return a shortlist for your laboratory to test.
Real coordinates, rendered live from PDB 3HFM (an antibody Fv bound to lysozyme). Neither molecule was computationally designed, and the ordering step is a conceptual illustration, not a model output. Drag to turn.
What happens. Design is conditioned on the target structure and on the patch of it a binder should engage.
Why it matters. De novo design starts with no antibody to improve, so the target surface and the objective are the only things anchoring the work.
May produce. A design brief: target, region of interest where known, antibody format and constraints.
What happens. Candidate backbones are explored against that surface, working from a disordered representation towards an ordered one.
Why it matters. Generation searches for a shape that could sit against the chosen surface. It is a search, not a test.
May produce. Candidate backbones with proposed sequences, each with an identifier.
What happens. A candidate backbone stands against the target. A sequence still has to fold to it, and the pair still needs computational assessment.
Why it matters. A shape that fits the target is a reason to look further, not evidence that it binds.
May produce. Candidates to carry into redesign, complex prediction and finally your laboratory.
Holding on “Target”. Select another stage, or play through them.
Three different things, kept apart
De novo design produces the first two. Only an experiment produces the third, and nothing on this page substitutes for it.
- 1
A designed sequence
A candidate the generation step proposed. It is a sequence predicted to fold into a plausible variable domain, nothing more.
Comes from · Generation and sequence redesign
- 2
A predicted structure
The same candidate folded in complex with the target, with a model-confidence estimate for the interface. Confidence is the model's own, not a probability of success.
Comes from · Cofolding against the target
- 3
A confirmed binder
A candidate shown to bind, with measured affinity and specificity. No amount of computational confidence establishes this.
Comes from · Your laboratory — SPR, BLI, ELISA and specificity panels
From Inputs to Insights
- Target antigen sequence, and a structure if you have one
- The research objective, for example blocking an interaction or targeting a conserved region
- A known or hypothesized epitope or region of interest, where one exists
- Antibody format: single-domain VHH or paired VH/VL
- Framework and developability constraints to respect
- Reference antibodies or prior discovery data, if any exist
Design can also start from a broader surface and narrow later. Not every item is required.
Explore related services
- Single Lead OptimizationEvaluate one existing antibody lead and explore variants of it.
- Lead OptimizationCompare and prioritize several candidates toward an objective.
- NGS Data AnalysisTranslate, group and count sequences from sequencing data.
- Epitope IdentificationExamine potential binding regions on a target.
No antibody yet? Start here.
Tell us about your target and objective. We will discuss whether and how a design workflow fits.