Private beta · YC S26
Now in beta · Request access

A thousand seeds.One binder.

Senbazuru — fold a thousand, grant one wish.

Twelve protein-design tools, one workflow. Point at a target, route across RFdiffusion, BindCraft, Boltz-2 and the rest, and rank survivors under a single rubric.

No credit card · Free tier · Cancel anytime
12tools, one rubric1,000seeds per wish3pre-registered targetsAUC 0.806ipSAE on holdout1API key0Conda envs12tools, one rubric1,000seeds per wish3pre-registered targetsAUC 0.806ipSAE on holdout1API key0Conda envs
The roster
Binder design
RFdiffusionRFdiffusionAAChromaBoltzGenBindCraftGerminal
Sequence design
ProteinMPNNLigandMPNNSolubleMPNN
Structure prediction
Boltz-2Boltz-1Chai-1AlphaFold 3
Embeddings & scoring
ESM-2ESM-C
From the benchiGEM 2025 finalist teamYC S26 candidateBuilt alongside three protein labs
§ 01 — The Problem

Designing a binder shouldn’t feel like gambling.

01

Every tool claims SOTA. None of them agree.

ProteinMPNN wins on one target, RFdiffusion on the next, BindCraft on a third. There's a right tool for your target — but nobody's measured which.

02

A thousand seeds is a weekend of shell scripts.

Conda conflicts. CUDA mismatches. Silent failures at seed 847. You came to design proteins, not maintain containers.

03

Every benchmark is a different ruler.

Each tool reports on its own target with its own scoring. We run every major tool against the same three targets, with one rubric and pre-registered thresholds — and publish the full dataset.

§ 02 — How it works

Four folds, then a crane takes shape.

From PDB to ranked candidates in a single run — no Conda, no orphaned shell scripts, no silent failures at seed 847.

I

Upload your target.

Drop a PDB, specify a hotspot — or let us infer one. We parse topology, surface, and interface class in under a second.

Input · PDB/CIF
II

We allocate the seeds.

Our routing algorithm reads your target — helical bundle, β-sheet, disordered loop — and distributes budget across the tools most likely to bind your geometry.

Allocator v1 · validated ipSAE AUC 0.806
III

Everything runs in parallel.

BYO compute on Free. Managed GPUs on Pro. Dedicated cluster on Frontier. No Conda. No CUDA. No seed 847.

Modal · Vast.ai · your VPC
IV

Ranked candidates, order-ready.

Every design scored by ipSAE, DockQ, pTMEnergy. Download the top 10, export to Twist, or push straight to a wet-lab partner.

Output · Scored PDB/FASTA
PDB→ Allocator→ GPU cluster→ Boltz-2 scorerRanked binders
§ 03 — The Catalog

Every tool, version-pinned, containerized, benchmarked.

A curated index of every protein design tool worth running — with reproducible environments, pinned weights, and head-to-head metrics on the same targets.

2023

RFdiffusion

Generative backbone
Baker LabOpen
2022

ProteinMPNN

Sequence design
Baker LabMIT
2024

BindCraft

End-to-end binder
Pacesa et al.Open
2024

Chai-1

Structure prediction
Chai DiscoveryClosed · Frontier
2025

Boltz-2

Validator
MITOpen
2023

Chroma

Generative model
Generate:BioOpen
2024

LigandMPNN

Ligand-aware design
Baker LabOpen
2024

AlphaFold 3

Structure prediction
IsomorphicClosed · Frontier
+ 4 more in catalogBrowse the full catalog
§ 04 — The Benchmark· PRE-REGISTERED ·

We publish the methodology before the results.

Targets, hotspots, seed counts, hit thresholds, and statistics are declared up front in a signed, versioned specification, so no threshold can be chosen after the fact.

Runs are not published yet. The leaderboard opens when a run set reproduces from a clean environment against those pre-registered criteria — not before.

Scoring stackBoltz-2 · ipSAE · DockQ · Rosetta ΔΔG
Hit criteriaipSAE ≥ 0.35 · ipTM ≥ 0.60 · pLDDTᵢ ≥ 70
Statistics95% Wilson CI · paired bootstrap
OversightExternal advisory panel · public comment
Paper Crane Benchmark · v1.0
Pre-registered. Results pending.
· specification published ·

The protocol is public and frozen. The runs are not published yet — the execution pipeline is still being validated end-to-end, and we will not show a number we cannot reproduce from a clean environment. The leaderboard appears here the day the first run set clears that gate, against thresholds that were written down long before we saw them.

Pre-registered
Targets, hotspots, seed counts, hit thresholds, and statistics were fixed and published before any run was scored. Nothing can be tuned after seeing results.
Signed & versioned
Methodology v1.0 is git-signed and immutable after publication. Changing a threshold, target, or statistic forks a new version with a public diff.
External advisory panel
Three external members — structural biology, experimental protein engineering, and ML methodology — review methodology changes and adjudicate reproducibility disputes.
7-day public comment
Every proposed new target goes through a seven-day public comment window before the panel votes on admission.
§ 05 — Pricing

Three folds. No surprises.

Free tier is free, forever. Pro is a flat subscription plus metered compute. Frontier is annual, invoiced, with a named account lead. That’s it.

Free
$0forever
For iGEM teams · students · academics
  • Every open-source tool
  • Bring your own compute
  • Public leaderboard access
  • Community support
  • 7-day data retention
Start free
Most chosen
Pro
$299/ month
For Labs · core facilities · biotech
  • Managed GPU, metered
  • 50,000 seeds / month
  • Private campaigns
  • Priority queue
  • Email support · 24h
  • 90-day retention
Start 14-day trial
Frontier
$50K+/ year
For Pharma · Series B+ biotech
  • Closed frontier models
  • Dedicated cluster
  • VPC or on-prem deploy
  • Custom tool integration
  • Shared Slack · 1h SLA
  • SOC 2 · HIPAA eligible
Talk to us
§ 06 — Built by

Two high-schoolers who got tired of Conda.

We’re Harsha Poonepalle and Vedant Kalipatnapu — iGEM co-captains and international winners at the iGEM Grand Jamboree. We spent a full season running ProteinMPNN, RFdiffusion, and BindCraft against the same target, by hand, one seed at a time. We lost three weekends to environment conflicts before we shipped a single design that worked.

We built Paper Crane because we wanted the tool we couldn’t find. Advised by Dr. Bryan Bryson (MIT Biological Engineering), Dr. Metzker (ex-MIT/Harvard, bio founder), plus faculty at UCLA and UPenn. Our early users include iGEM teams across four continents.

We believe the next generation of therapeutics will be designed, not discovered — and that the platform underneath should be as elegant as the proteins it builds.

MIT · faculty advisory
UCLA · faculty advisory
UPenn · faculty advisory
“Fold enough paper, carefully enough, and one of them will heal something.”
— Paper Crane manifesto
Founded
2026
HQ
Atlanta → SF
Batch
YC S26
§ 07 — Questions

The answers to the questions.

Free is free forever on your own hardware. Pro is a flat monthly subscription plus metered compute (pennies per seed). Frontier is annual, invoiced, with a dedicated account manager.

Every major open-source tool: ProteinMPNN, RFdiffusion, BindCraft, Chroma, Genie2, LigandMPNN, ESM-IF, BoltzDesign, plus validators (AF3, Boltz-2, Protenix, Chai-1). Closed partners available on Frontier.

Yes. On Pro, your inputs and outputs are encrypted at rest and never used for training. On Frontier we deploy into your VPC — we never see your data. Free tier runs on your own hardware, so privacy is yours.

Yes. Open a PR on our catalog repo with a Dockerfile and a benchmark run. We review weekly. Frontier customers get white-glove integration of proprietary tools.

Pro and Frontier are SOC 2 Type II (in progress) and HIPAA-eligible. All traffic TLS 1.3, storage AES-256, compute in sandboxed containers. Frontier deploys within your cloud account.

No. Your designs are yours. We don't own them, license them, or patent them. We're a platform, not a partner on your IP.

Private beta opening · Spring 2026

A thousand seeds. One binder.
Start folding.

Join the iGEM teams, academic labs, and biotechs already on the waitlist. No credit card. Free forever on your hardware.

Opens your email clienthello@papercrane.bio