Protein design proposes sequences and structures against specified constraints. Binder design adds a target and interaction objective. Start with the intended function, molecular context, and experimental validation plan.
Define the design brief
Specify the target construct, binder modality, epitope or pocket, protected residues, length range, desired properties, and candidate budget. Distinguish a completely new scaffold from redesign of an existing sequence or interface.
For enzymes, catalytic geometry and essential cofactors may need protection. For antibodies, framework, CDR definitions, and chain pairing matter. A peptide, nanobody, and full antibody are not interchangeable design inputs.
Select a generation strategy
| Objective | Tools to investigate | Review before running |
|---|---|---|
| Joint sequence/structure generation and contextual design | RFD4-Proteina | Supported mode, retained molecular context, and generated-file interpretation |
| Target-conditioned binders | BoltzGen, NeuroBind, BindCraft2 | Binder modality, target selections, sampling, and downstream validation |
| Antibody-specific generation | RFantibody | Framework and CDR constraints and antigen context |
| Sequences for an existing structure | LigandMPNN | Fixed positions, ligand/cofactor context, and sequence diversity |
| Enzyme optimization | NeuroFold2 | Property objectives, protected functional regions, and supported system |
Check the current service form; availability of one mode does not imply support for every molecular system. Generated structures are proposals. In particular, do not treat generation-only outputs as independently refolded or affinity-ranked results.
Evaluate candidates independently
Refold selected sequences with an appropriate structure predictor, compare intended versus predicted folds, and inspect interfaces. Score developability using methods suited to the modality.
Preserve the distinction between generator confidence, foldability, interface confidence, predicted affinity, and measured binding. Reusing the generator’s own score is not independent validation.
Select a bounded experimental set
Apply declared filters, review diversity, and retain failure information. Save target versions, selections, seeds, designed sequences, candidate IDs, structures, and native scores. A library of near-identical top-scoring designs may test fewer hypotheses than a diverse set supported by several types of evidence.
Use bulk planning before multiplying candidate counts through folding and scoring. See protein design solutions for related applications.