Rank small-molecule candidates by predicted binding affinity
BindBench is a screening workbench for small biotech labs. Point it at a protein target, feed it a compound library, and get a ranked shortlist powered by the Boltz-2 model, backed by Postgres with pgvector.
Register protein targets with sequences and UniProt IDs. Build compound libraries by hand or bulk CSV upload, stored with vector embeddings for similarity search.
Boltz-2 affinity predictions
Fan out one prediction job per compound against the Boltz-2 API, then poll and rank as results land. A deterministic mock mode runs the whole loop with zero API cost.
Ranked, reviewable results
Every screen produces a ranked table of predictions with affinity scores, confidence metrics, and per-compound status from queued to succeeded.
How it works
01
Create a project
Name the campaign and attach a protein target with its sequence.
02
Add compounds
Upload a CSV of SMILES strings or add candidates one by one.
03
Run the screen
Start prediction jobs for the full library in one click.
04
Review the ranking
Compare predicted affinities side by side and export the shortlist.
Built with
Next.js 16
TypeScript
Tailwind CSS
Drizzle ORM
PostgreSQL + pgvector
Boltz-2 API
Vercel Cron
See it in action
The live deployment ships with a demo project so you can explore a finished screen immediately.