Tutorials

1. Design your first binder

About 10 minutes of setup, then GPU time. Nothing is charged until you press submit, and the cost is shown before you do.

1. Load a targetType a PDB ID into the search bar above the viewer — 5C3T is a good first target — and pick the biological assembly. The structure loads into the 3D view with its chains listed.
2. Trim to the domain you care aboutKeep only the chains and residue ranges involved in the interaction. This is the biggest lever on both runtime and cost: the cost curve rises with the square of residue count above a per-tool knee. The structure is renumbered for you.
3. Mark hotspotsClick the residues you want the binder to engage, in the 3D view or on the sequence track. If your target is already in complex with a partner, use interface detection to propose the set and then edit it.
4. Pick a toolStart with BindCraft. Set the number of designs — 20 to 50 is a reasonable first campaign — and leave the advanced filters at their defaults.
5. Check the quote and submitThe studio shows per-design and total cost. Tick “email me on completion” if the run is long, then submit. Jobs are dispatched to GPUs that cold-start from zero, so the first result can take a while to appear.
6. Screen the resultsSort by ipTM, then check hotspot coverage — a high-confidence binder that bound the wrong face is not a hit. Click any row to load its structure in the viewer next to the target.

2. Fold a sequence you already have

1. Switch to Fold modeOpen New Experiment and choose Fold. There is no target/binder distinction here — you are just predicting a structure.
2. Paste or upload your sequencesPaste a raw sequence, or drop a FASTA file. Add one entity per chain to fold a complex rather than a monomer.
3. Choose speed or accuracyESMFold2 returns a single-sequence prediction quickly. Boltz-2 uses a cached target MSA and is the better call when you need confidence metrics you intend to act on. Protenix handles ligands and nucleic acids.
4. Inspect and shareOpen the result in the viewer, colour by confidence, and use the share button to hand someone a link that opens the structure without a login.

3. Refine an existing design

1. Select a design from a finished runAny design with a binder sequence can be used as a seed, regardless of which tool produced it.
2. Reseed into ProteinHunterThe design action opens a ProteinHunter run in refine mode, prefilled with that design's target and binder. It searches outward from your starting sequence in cycles.
3. Compare across cyclesProteinHunter reports per-cycle metrics, so you can see whether the search is still improving or has plateaued before spending more GPU time.

Practical notes

• Trim before you scale. Halving residue count more than halves cost. Run a small campaign on a trimmed target before committing to a large one.
• Screen on the interface, not the fold. pLDDT tells you the model is confident about the shape; ipTM and ipSAE tell you it is confident about the binding.
• Watch for single helices. An asphericity above ~0.45 means the binder is essentially a rod. These often score well and behave poorly.
• Group related runs into a project. Experiments in a project share a target and can be compared side by side.

Want one run with us?

If you would rather have the first campaign designed with you than work through it alone, we do that.