Drug discovery, from a target protein to ranked compounds, in your browser
Talindrew's AI workspace runs structure-based drug discovery as a visual pipeline: fold the target, find its pockets, dock a compound library, filter by predicted ADMET, check poses and selectivity, rank, and design analogs for the next round. Every stage shows the model that ran and what it found, and any stage can use your own model, including AlphaFold 3 under your licence. It sits next to the plasmids, primers and notebook entries the target came from.
What it does
Structure prediction
ESMFold in seconds, Boltz-2 for protein-ligand co-folding with affinity, a structure you bring, or your own model such as AlphaFold 3 under your licence.
Which model to usePocket finding
P2Rank or fpocket find druggable cavities, ranked, with centres ready to dock into.
Binding pocket predictionDocking and screening
AutoDock Vina or GPU Uni-Dock over approved drugs, fragments, natural products or your own SMILES.
Virtual screening, step by stepADMET, pose checks, selectivity
ADMET-AI predictions, PoseBusters validity checks, Boltz-2 rescoring and an off-target panel, weighed into one ranking.
The FoldRx pipelineIterative design
CReM proposes analogs of each round's best hits; the next round screens them, reusing every stage that did not change.
How rounds workYour own models
Any stage can call an endpoint you run: AlphaFold 3, ESMFold2, a fine-tuned model or an NVIDIA NIM, for your account only.
Use your own modelGuides
- Use your own model in Talindrew, including AlphaFold 3
Plug your own model into any FoldRx stage: the request and response contract, 202 polling, tokens, and a one-file adapter that serves your licensed AlphaFold 3 to Talindrew.
- Can you use AlphaFold 3 commercially?
AlphaFold 3 weights are non-commercial only; the code is Apache 2.0. What the terms allow, AlphaFold Server's 30 jobs a day, the Google Cloud route, and the open models you can use commercially.
- ESMFold2 vs AlphaFold 3 vs Boltz-2
How ESMFold, ESMFold2, AlphaFold 3 and Boltz-2 differ: MSA or single sequence, ligands and complexes, speed, confidence scores and licences, and which to use in a drug discovery pipeline.
- Structure-based virtual screening, step by step
A structure-based virtual screening workflow in order: target structure, pocket prediction, compound library, docking, ADMET filters, pose checks, selectivity and ranking, with the tools for each step.
- Binding pocket prediction: P2Rank vs fpocket
How ligand binding site prediction works, P2Rank's machine-learned scoring versus fpocket's alpha spheres, how to choose the pocket to dock into, and what changes on predicted structures.
Comparing platforms? Talindrew vs Neurosnap · Talindrew vs Tamarind Bio
Drug discovery questions
What is Talindrew for drug discovery?
Talindrew is a browser workbench for molecular biology whose AI workspace runs structure-based drug discovery: it folds a target protein, finds its pockets, docks a compound library, filters by predicted ADMET, checks poses and off-target selectivity, ranks the candidates and designs analogs for the next round. Each stage shows which model ran and what it found.
Do I need to code or have a GPU cluster?
No. Pipelines are built visually from stage nodes in the browser, and the models run on Talindrew's GPUs. Developers can also plug in their own model endpoints.
Can I use AlphaFold 3?
Talindrew cannot run AlphaFold 3 for you, because its weights are licensed for non-commercial use only. If you hold the licence, run it yourself behind Talindrew's open-source adapter and add it as your own model, or upload a structure from AlphaFold Server. Boltz-2, built in, is an MIT-licensed AlphaFold 3-class alternative.
How much does it cost?
Cloning, sequence editing and the notebook are free. Drug-discovery runs use GPUs, so they are on Pro (10 runs a month from $24) and Max (50 a month); finished demo screens are open to everyone.
Open a finished screen in the AI workspace.
Demo screens are free to explore. Runs are on Pro and Max.