FAQ · 19 questions6 CATEGORIES
Questions, folded once.
Short answers about beta access, supported tools, compute, scoring, privacy, and reproducibility. If something here is wrong or missing, write to hello@papercrane.bio.
Beta access
3 Q- Q01What's in the beta?
- A hosted workflow surface for the binder-design pipeline we run for our own campaigns: target intake, tool orchestration (RFdiffusion, ProteinMPNN, BindCraft, BoltzGen), structure prediction (Boltz-2, Chai-1, AlphaFold2), and ranking with ipSAE / iPTM / pDockQ. No wet-lab automation, no LIMS.
- Q02How do I get access?
- Request an invite from the homepage or email beta@papercrane.bio. iGEM teams and academic labs are prioritized while the beta is capacity-limited.
- Q03Who is the beta for right now?
- Computational protein-design groups with a defined target and the ability to evaluate predictions. We are not a fit yet for teams without a structural-biology background.
Tools
3 Q- Q04Which tools can I run?
- RFdiffusion, ProteinMPNN / SolubleMPNN / LigandMPNN, BindCraft, BoltzGen for design; Boltz-1/2, Chai-1, and AlphaFold2 / AF-Multimer for prediction; Foldseek and ESM2 for retrieval and scoring; ipSAE for ranking. The set tracks what we use internally.
- Q05Can I add a tool that isn't on the list?
- During beta, custom tools land case-by-case. If it runs in a container and accepts a sequence or structure, it can usually be wired in. Tell us what you need.
- Q06Do you support multi-step workflows?
- Yes. A run is a DAG of steps, each step is a single tool, and outputs of one step are addressable inputs to the next. The same graph executes locally, on Modal, or on Vast.ai.
Compute (BYOK Modal)
3 Q- Q07Do I need my own GPU?
- No. Default execution uses a shared backend with metered usage. If you want full control over cost and quotas, bring your own.
- Q08Can I bring my own Modal account?
- Yes. Paste a Modal token and runs execute under your workspace - your quota, your billing, your logs. We never proxy your token. The same pattern works for Vast.ai instance keys.
- Q09What happens if a job fails mid-run?
- Each step is idempotent and content-addressed. Reruns skip steps whose inputs haven't changed; failures surface with the underlying tool's stderr, not a wrapped error.
Scoring
3 Q- Q10How do you score binders?
- Predictions are run through Boltz-2 or Chai-1 (and AF2 when requested), then ranked by a composite of interface metrics. ipSAE is the primary signal; iPTM and pDockQ are reported alongside. Thresholds are user-configurable per campaign.
- Q11What's ipSAE?
- ipSAE: interprotein Score from Aligned Errors: is a binder-ranking metric derived from the predicted aligned-error matrix at the interface. We surface it because, in our hands and in the literature it draws on, it discriminates real interfaces better than iPTM alone.
- Q12Do you publish your filtering thresholds?
- Yes. Default thresholds and the rationale behind them live in the docs. You can override any of them per run.
Data privacy
4 Q- Q13Where is my data stored?
- Sequences, structures, and run artifacts are stored in object storage scoped to your workspace. Logs and metadata sit in our primary database. See the privacy policy for region and retention details.
- Q14Do you train on my designs?
- No. Customer designs are not used to train models.
- Q15Can I export my designs?
- Yes. Every run exports as a folder of FASTA, PDB / CIF, JSON metrics, and the full run manifest. There is no proprietary file format and no export wall.
- Q16Can I delete my data?
- Yes. Workspace deletion removes artifacts and metadata; backups age out per the retention schedule in the privacy policy.
Reproducibility
3 Q- Q17What's the public benchmark?
- A pre-registered comparison of design pipelines on a fixed set of targets, scored with the same metrics and seeds for everyone. The methodology is published and frozen before any result is scored; runs are not published yet. See /benchmark for the specification and release criteria.
- Q18Are runs reproducible?
- Each run pins tool versions, container digests, seeds, and inputs into a manifest. Re-running the manifest on the same backend reproduces the artifacts; switching backends reproduces the sequences and structures up to nondeterminism documented per tool.
- Q19Do you offer pricing for academic labs?
- Yes. Free for iGEM teams and academics. Lab and pharma tiers are on the pricing page.