Binder Design Lab
Design protein binders against any target. RFdiffusion, BoltzGen, BindCraft, and ProGen2 — one API key, ranked candidates, predicted ΔG affinity, ready-to-order sequences.
Featured models
What's inside
RFdiffusion + RFdiffusion2
The viral protein-design model. Every 'I made a binder this weekend' tweet runs through it. RFdiff2 adds enzyme active-site scaffolding with natural-level catalytic proficiency.
BoltzGen Universal Binder
The 2026 successor to RFdiffusion. Sequence-only input, no target PDB needed. Returns candidate binders with predicted ΔG binding affinity and confidence scores.
BindCraft End-to-End Pipeline
Chains BoltzGen → Boltz-2 validation → ProteinMPNN refinement → composite ranking in one API call. Published wetlab hit rates of 10-100%. Save days of orchestration code.
ProGen2 Conditional Generation
Generate novel protein sequences conditioned on family, EC number, or starting motif. Naturalness scoring, length control, and family-aware priors.
Composite Ranking
Every BindCraft candidate gets a composite score combining Boltz-2 ipTM, ProteinMPNN recovery, ΔG affinity, and predicted wetlab success. Ready-to-order candidates.
Drop-in Lab Integration
Every endpoint returns PDB files, interface residue lists, and per-design confidence scores. Direct integration with Twist, IDT, or GenScript for wetlab ordering.
FAQ
What's the difference between RFdiffusion and BoltzGen?›
RFdiffusion (Baker Lab, 2023) operates on protein backbones — it requires a target PDB structure and generates new backbones. BoltzGen (MIT Jameel Clinic, 2026) operates on sequences directly — you just pass a target amino-acid sequence and it generates candidate binder sequences with predicted ΔG. Use RFdiffusion when you have high-quality target structures; use BoltzGen when you only have sequences or want faster turnaround.
How accurate are these designs?›
All four models are published, peer-reviewed methods. BindCraft reports 10-100% wetlab hit rates depending on the target. BoltzGen and RFdiffusion2 have demonstrated real experimental validation in multiple campaigns. That said: these are research tools, not clinical-grade predictions. Always validate your top candidates experimentally before committing to larger campaigns.
Can I download the PDB files of designed binders?›
Yes. RFdiffusion and BindCraft both return PDB file contents in the API response. You can save these and open them in PyMOL, ChimeraX, or Mol* for visualization. BoltzGen returns sequences and binding metrics; feed those through ESMFold or Boltz-2 to get structures.
What's the cost per design?›
RFdiffusion is $0.12/design (12 credits), BoltzGen is $0.15/design (15 credits), the full BindCraft pipeline is $0.40 per run (40 credits, returns ~5 ranked designs). ProGen2 is $0.08/sequence (8 credits). Free tier users get 500 credits/month — enough for a couple of binder design campaigns.
Are the real models running, or is this mock data?›
SciRouter ships with a production-ready API surface and deterministic mock mode for testing. Real RunPod GPU workers for RFdiffusion, BoltzGen, and BindCraft are being deployed incrementally; until each endpoint goes live in real mode, responses come from a deterministic mock that returns plausible structures and scores. Every response includes a 'dispatch_mode' field (mock | runpod) so you always know which backend served your call.
Is this a clinical tool?›
No. This is a research and design tool. Designed binders require extensive wetlab validation, safety testing, and regulatory review before any clinical use. SciRouter provides the computational design stack — the wetlab and clinical stack is still up to you.
Ready to design a binder?
Open the lab, paste a target, and start designing.