the platform
fold a protein. dock a ligand. plan the synthesis.
the computational stages of pre-clinical work, from target identification to synthesis planning — 500+ tools that hand back a record, not an answer you have to take on trust.
sequence → synthesis
from target to synthesis route, in one place.
every stage runs in the cloud, on GPUs where the model needs them, with no local setup, and you pick the model that drives it: anthropic direct or through openrouter, across the model families your plan includes — or bring your own key once you sign in.
- sequence
start from a target — a FASTA sequence or a named protein.
- structure
fold it to a predicted structure (ESMFold).
- dock
dock ligands against it (DiffDock), searched or generated.
- admet / tox
screen the hits for ADMET and toxicity (ADMET-AI).
- synthesis
get a retrosynthetic route for the ones you keep (AiZynth), then look up vendors (ZINC22).
asked often
what people ask before their first run.
ivy biosciences is a platform for computational biology, FSL-1.1-Apache-2.0 licensed with a source release planned. it provides 500+ tools covering protein structure prediction (esmfold) and protein design (esm3), molecular docking (diffdock), ADMET and toxicity screening (admet-ai), and generative molecular design (genmol). Ivy chains these tools in one question, from target identification through synthesis planning, each job's output feeding the next, without any infrastructure to set up.
drug discovery is one application: bringing a drug to approval takes 10-15 years and averages $2.6 billion per approved drug (Tufts CSDD, 2016). computational screening evaluates candidates virtually in hours, predicts protein structures without wet-lab crystallography, and estimates ADMET properties before synthesis — so the wet-lab work that follows starts from better-ranked candidates.
yes. a guest run needs no account or card, and the free Seed tier includes 500 credits per month with no card required. students with a verified .edu email get the student plan: 3000 credits a month and Grow-equivalent workflow access, no payment. paid plans start at $20/month — see /pricing for the full list.
the tools are the same ones you would install — esmfold, diffdock, admet-ai, aizynthfinder. what changes is that one record spans the chain: a structure predicted by esmfold can be docked by diffdock and screened by admet-ai without you moving files between environments or rebuilding each dependency. running them yourself gives you the same models and more control; chaining five stages by hand costs the setup time and the provenance trail between steps.
a guest run is free up to $1.00 of compute a day with no account or card, and an unsaved guest run is not kept. the free Seed tier includes 500 credits a month, and a verified .edu student plan includes 3000. paid plans start at $20/month. every tool shows its credit cost before the job is submitted, so a run cannot bill you by surprise.
the platform includes 500+ tools from target to synthesis route: protein structure prediction, molecular docking, ADMET prediction, generative molecular design, retrosynthesis planning (aizynthfinder), binding site detection (p2rank), single-cell analysis, and literature mining (biobert). every tool runs in the cloud, on a GPU where the model needs one, with no local setup required.
ready when you are
start a run. keep the record.
chain models into a workflow from whichever surface you already work in — and leave with the artifact. every result names the tool, parameters and run that produced it. no setup required.