Start with the Target. Design New Antibodies.
Explore epitope-targeted de novo antibody design through computational structure generation, sequence redesign and complex evaluation. De novo design starts without an existing antibody lead.
Binder navy, its CDR3 region gold, target sand, interface coral. Drag to rotate; press an i for a term. Structure: PDB 3OGO (RCSB PDB), a single-domain nanobody on green fluorescent protein. Real coordinates of an experimentally determined complex, not a designed molecule.
From target to candidate review
Three computational stages, with project framing before and after them. Scroll to follow each one; the picture keeps pace.
- 1. Define the target
- 2. Generate candidates
- 3. Redesign positions
- 4. Predict the complex
- 5. Compare and hand off
- Solid: computational · dashed: project framing
- Stage 1Project framing
Define the target
Agree what the binder should engage, and in which format.
- In: target, region of interest, format
- Method: project framing, not an automated step
- Out: a defined design task
Limit. Framing the task does not show that a binder can be designed against this surface.
Region of interestReal target coordinates (green fluorescent protein, PDB 3OGO), shown with its binder hidden. The three regions are an even split of its surface, drawn to show how a region is chosen; they are not computed epitopes.
- Stage 2Computational
Generate candidates
Propose candidate binder backbones against the chosen region.
- In: target structure and design objective
- Method: structure-guided generative design (hallucination-based)
- Out: candidate backbones with proposed sequences
Limit. A search for structures that look plausible to a prediction network, not a binding assay.
A real single-domain binder backbone (PDB 3OGO, chain G), shown on its own as an example of what a candidate backbone looks like. Nothing here was generated, and a project proposes several, each different.
- Stage 3Computational
Redesign positions
Re-propose selected positions on a fixed backbone.
- In: candidate backbones
- Method: antibody-aware inverse folding
- Out: redesigned candidate sequences
Limit. Does not guarantee a candidate passes the next stage.
Schematic. Positions are numbered, not named: no amino acid appears, so nothing here can be mistaken for a real design. Which positions are chosen is set per project, commonly in the CDR loops and near the interface.
- Stage 4Computational
Predict the complex
Check whether a candidate and the target look structurally compatible.
- In: redesigned sequences and the target
- Method: complex structure prediction with interface confidence
- Out: predicted complexes, each with its confidence
Limit. Confidence is not affinity, specificity or function. Those need assays such as SPR, BLI or ELISA.
A deposited complex (PDB 3OGO, a single-domain binder on its target) standing in for what a predicted complex looks like; no prediction was run for this view. In a project, the dashed contacts would be predicted. Confidence is the model’s own statement of how sure it is about the placement, and is not a measure of binding.
- Stage 5Project framing
Compare and hand off
Choose what merits review, and name what an assay must answer.
- In: predicted complexes and the design brief
- Method: computational review, not a pipeline step
- Out: comparison files and the open questions
Limit. There is no single score. A shortlist is not a set of validated binders.
Illustrative placeholder comparison of three candidates. Press a row to see what would be examined. No values, scores or rankings are shown. Illustrative candidate Sequence Predicted complex Assay question Reviewed Reviewed Open Reviewed To review Open To review To review Open Press a candidate to see what would be examined for it.Illustrative placeholder rows. They are not results, and no number, score or ranking is shown.
Real target coordinates (green fluorescent protein, PDB 3OGO), shown with its binder hidden. The three regions are an even split of its surface, drawn to show how a region is chosen; they are not computed epitopes.
Three different things
A designed sequence, a predicted structure and a confirmed binder are not the same, and they are never treated as the same on this page.
- 1Produced here
A designed sequence
The output of generation and redesign: a sequence proposed to fold into a plausible binder.
- 2Produced here
A predicted structure
The output of complex prediction: a model of the candidate against the target, with the model's own confidence estimate.
- 3Needs your laboratory
An experimentally confirmed binder
Only an assay such as SPR, BLI or ELISA can show that a candidate binds. IndiskaAI does not run wet-lab validation unless that is separately confirmed and scoped.
A shortlist is a set of candidates that merit further review, not a set of validated binders. Nothing on this page is a success rate, an affinity, a benchmark or a client result.
From Inputs to Insights
Research Inputs
The inputs the workflow needs.
Scientific Analysis
Three computational stages, described by what they do.
Research Outputs
Press an item to read it.
Every item is within agreed scope. A project can return no candidates that pass its filters.
Computational outputs support prioritisation. They do not prove binding, specificity, potency, safety or developability, and experimental validation is still required.
Have a target or epitope in mind?
Tell us about your target and what you want to engage, and we will discuss whether and how this workflow fits.