LIGAND
Every wallet is a
candidate molecule.
Your Robinhood Chain wallet picks a real natural compound and docks it against fourteen cancer targets.
From wallet to binding map
The block molecule
Every new block hash picks a molecule too.
Today's strongest binds
Good questions
Find, dock and recombine your molecule
Your address picks a real natural compound. Map it, dock it, then cross it with other wallets.
Your molecule
The active address is hashed with SHA-256. Bits 0–31 of that fingerprint pick one of the natural compounds; the same address always gets the same molecule.
----
Molecular fingerprint
SHA-256 · 256 bitsAddress
inputBinding map
Your molecule is docked against all fourteen targets with a quick search. The strongest target is carried into step 3. Computational estimate only.
Binding map
14 targets · computational estimateAll targets
kcal/mol| # | Target | Cancer | Score | H-bonds | Band |
|---|
Quick rigid docking with the in-browser Vina-style engine: one PubChem conformer, 4 Monte Carlo runs per target, seeded by the fingerprint so the same address always gives the same map. These are computational estimates for ranking only. They say nothing about activity in cells, animals or people.
Strongest target
after the mapThe best target appears here after the binding map: its score, the docked pose in 3D and the key contacts.
Strongest target
----
Dock in detail
The full workbench opens on your molecule and its best target. Tune the search, inspect poses, score terms, interactions and the alanine scan. Approved drugs stay available as a reference.
Idle
0%Convergence
best score per step| # | Score | RMSD Å | H-bonds | Hydrophobic residues | Assessment |
|---|---|---|---|---|---|
| Run docking to see poses. Click any row to view it in 3D. | |||||
Run docking or score the crystal pose.
Score = (sum of pairwise terms) ÷ (1 + 0.0585 × rotatable bonds). For every protein–ligand heavy-atom pair within 8 Å, the surface distance d is fed into five terms:
| gauss 1 | exp(−(d/0.5)²) | −0.0356 |
| gauss 2 | exp(−((d−3)/2)²) | −0.00516 |
| repulsion | d² when d < 0 | 0.840 |
| hydrophobic | both atoms hydrophobic | −0.0351 |
| hydrogen bond | donor–acceptor pair | −0.587 |
Weights from Trott & Olson (2010), the AutoDock Vina paper. This is an independent implementation, not AutoDock Vina itself.
Run docking or score the crystal pose.
Each pocket residue is mutated to alanine in turn (side chain removed beyond Cβ) and the current pose is rescored. The larger the penalty, the more that residue contributes to binding.
Run docking or score the crystal pose first.
Compare & recombine
Dock another wallet's molecule into the same target, or mix the two fingerprints into a child molecule and see whether it beats both parents.
Your molecule against another wallet
Both molecules are docked into the selected target with the same exhaustiveness, box and seed. Recombination mixes the two 256-bit fingerprints into a child fingerprint, which selects a child molecule from the same library.
Target library
Fourteen co-crystal structures of approved targeted drugs. Open any row for the 3D structure, binding pocket and sequence.
| Target | Drug | Indication | PDB | Resolution Å | Approved | Atoms | Crystal score |
|---|
All experiments
Every bind is recorded, including the ones that failed.
| Time | Wallet | Target | Ligand | Score | Block | Status |
|---|
Why publish failures
A weak bind is still information: it tells the next person which path does not work. LIGAND keeps unconverged and weak runs in the same ledger as the strong ones.
Score distribution
last 60 runsTop contributors
Ranked by docking runs contributed and by best bind per target. Run one docking job and you will appear here.
Contributors
runs| # | Wallet | Runs | Best | Failures |
|---|
Best bind per target
kcal/mol| Target | Ligand | Score | Wallet |
|---|
Methods & mechanisms
Behind every target is a chapter in the history of targeted therapy. Educational summaries of public knowledge, not medical advice.
How it is built
Architecture
LIGAND has four layers: data (protein–ligand co-crystal structures from the RCSB Protein Data Bank), compute (a scoring and docking engine that runs in the browser), result (target, ligand, parameters and score hashed together), and record (the hash bound to a Robinhood Chain block).
All computation happens on the visitor's device. The page does not depend on any private server.
Scoring function
For every protein–ligand heavy-atom pair within 8 Å, the surface distance is d = r − R₁ − R₂ using van der Waals radii (C 1.9, N 1.8, O 1.7, S 2.0, F 1.5, Cl 1.8 Å). Five terms are summed: gauss 1, gauss 2, repulsion, hydrophobic and hydrogen bond, weighted −0.0356, −0.00516, 0.840, −0.0351 and −0.587. The sum is divided by 1 + 0.0585 × the number of rotatable bonds.
Docking precomputes one energy grid per ligand atom type (0.45 Å spacing), then runs independent Monte Carlo searches with local optimisation. Each run's best pose is rescored exactly and poses are clustered at 2 Å RMSD.
Approximations:
- Rigid docking: the ligand keeps its crystal conformation.
- No hydrogens: donors and acceptors are assigned from element and connectivity.
- Rotatable bonds are estimated from bond lengths and ring membership.
- Only the ligand's protein chain is kept; waters and metal ions are removed.
- Covalent drugs (osimertinib, ibrutinib, sotorasib) show atom overlaps with the cysteine they bond to.
- The alanine scan deletes side chains and rescores; it does not re-optimise the pose or the protein.
Validation: with default settings (exhaustiveness 8, 22 Å box) and a fixed seed, 13 of 14 approved drugs were re-docked to their crystal pose within 1 Å. Over 10 random seeds: imatinib 8/10, venetoclax 10/10, olaparib 10/10, ibrutinib 9/10, lapatinib 5/10. Lapatinib is large and flexible; higher exhaustiveness helps.
Data sources
All structures were downloaded from files.rcsb.org and checked by title and ligand residue name. Each page keeps the ligand's protein chain and one copy of the ligand.
Open source
Real vs demo
Real: atomic coordinates, resolution, organism, deposition date, sequences and binding-pocket residues; every score, docking run, pose, RMSD, interaction list, alanine scan and property shown in the lab. The wallet connection is real and read-only: LIGAND asks your browser wallet (MetaMask, Rabby, OKX and other EIP-6963 wallets) for your address and may ask to switch to Robinhood Chain; it never requests a signature or a transaction.
Demo: other wallets' runs in the ledger, the optional demo wallet, block numbers and transaction hashes in simulated on-chain records, leaderboard entries and the overview statistics. Recording results on chain is coming soon; today nothing is sent from your wallet.
Robinhood Chain
Robinhood Chain (chain ID 4663) is used today for trading, tokenized stocks and memecoins. LIGAND shows another use: a public record of scientific computation. Each result hash maps to a block, and anyone with the same inputs can reproduce it with the same open tools.
Block numbers and transaction hashes on this page are demo data. Nothing is written on chain yet.
Disclaimer
LIGAND is a concept demonstration. Scores are computational estimates from a simplified empirical function and rigid docking. They rank candidates for further study and do not indicate activity in cells, animals or people. Nothing here is medical advice.