Pick a clinically relevant mutation, pick your compounds, and Liganx docks them against wild-type and the mutant in parallel — then ranks the shifts so you can prioritise what to test next. A fast first pass, not a final verdict. No PyMOL, no FoldX setup, no AutoDock wrangling.
Built for proprietary work. No account required, no third-party trackers or cookies, and nothing leaves North America. Your compounds are sent to a GPU worker only for the run itself, then discarded there — and we'll delete any job on request. How we handle your data →
| Compound | WT | T790M | C797S |
|---|---|---|---|
| Osimertinib | -8.8 | -9.3 | -5.2 |
| Gefitinib | -7.2 | -5.1 | -4.9 |
| Erlotinib | -7.6 | -5.4 | -5.3 |
| Compound X | -6.1 | -7.8 | -7.1 |
We publish the full pose-accuracy benchmark, the per-pose confidence-badge validation, and the per-target scores — so you can audit the numbers instead of taking a slogan on faith.
GNINA 69% top-1 < 2 Å on Astex-42, automated · confidence badge 12/13 right when green · see Validation →
Built on tools the community already trusts
Rank a target's known mutations by how likely each is to weaken your compound — in one click. No other self-serve tool packages this.
Bring your own molecule. In one pass, Resistance Radar docks it across a target's entire known resistance panel and returns a ranked liability map — per-variant Δ-vs-wild-type docking plus a forecast score for how likely each mutation is to confer resistance (calibration not yet validated). It's the resistance question asked about your compound, not a generic drug: where will it hold, and where will it fail?
Early-access beta. Panel coverage is growing target by target; request access from the Studio.
Choose from clinically actionable kinases or upload your own PDB. Pocket boxes are pre-defined.
Click EGFR T790M, KRAS G12C, BRAF V600E — anything from the curated library, or type your own.
We dock every compound against WT and each mutant. The Δ-score highlights how each compound's binding shifts between variants.
Every Boltz-2 pose now carries the model's own confidence (ligand pLDDT), binned green / amber / red and checked on a 30-target holdout of post-2024 structures Boltz-2 never trained on. There, green poses were right 12 of 13 times; low-confidence poses are flagged, so you don't chase a bad hit. A small-sample trust signal, not a calibrated probability.
For every FDA-approved targeted cancer drug, the Atlas ranks which mutations are most likely to emerge as clinical resistance. Combines docking Δ + ESM-2 protein-language-model fitness in a 2-signal logistic model, fit on 25 published clinical-resistance events (16 resistance / 9 non-resistance): ROC-AUC 0.90 in-sample, 0.81 5-fold cross-validated (95% CI 0.62–0.96 — small-n, indicative). 15 drugs covered today. Novel (gene, position, mutant) lookups now run real ESM-2 on our GPU pod on demand.
Upload up to 10 (gene, position, wt, mutant, drug) rows as CSV. We score each through the same 2-signal model the Atlas uses — real ESM-2 inference for novel mutations, instant cache hits for the 49-event ESM-2 cache — and return a forecast score (uncalibrated), verdict, and AUC if you provide ground truth. Free tier: 10 rows / day.
We pre-ran 30 oncology kinase inhibitors against every resistance mutation in our catalog — KRAS G12C/G12D/Q61H, EGFR T790M/L858R/C797S, BCR-ABL T315I/E255K. Hit a public URL and see ranked selectivity hits in 1 second. Click any compound to see its 3D pose.
N compounds × M mutants in one view, cells colored by Δ-score so resistance and selectivity gain pop out instantly. The whole product on one screen.
Single-snapshot rescore of the docked pose (ff14SB / OpenFF Sage / OBC2, ε=1, no salt). Ranks compounds within a target — not a Kd. Short MD is a pose-stability badge, never averaged in. Median Spearman ρ 0.71 across 8 congeneric series (184 ligands; range 0.46–0.83), rescoring reference poses. Available by request.
30 FDA-launched oncology kinase inhibitors pre-docked against every catalog resistance mutation — public landing pages, no login. Open any card on /library and see the ranked selectivity hits in one second.
Drop up to 1000 compounds against a (target, mutation) pair; we dock each against WT and the mutant in parallel and return a hit list ranked by selectivity index. Promote the top hits to a full job in one click — no re-dock.
A public per-drug atlas ranking the mutations most likely to break each FDA-approved targeted drug, fit on 25 clinical-resistance events (16 resistance / 9 non-resistance; out-of-fold ROC-AUC 0.81, 95% CI 0.62–0.96 — small-n). Upload your own (drug, mutation) CSV to score against the same ESM-2-backed model.
Free jobs run on QuickVina2-GPU, with every pose checked by PoseBusters, Vinardo re-score, and RDKit strain analysis — most free tools give you no validation at all. GNINA CNN rescoring and Boltz-2 ML co-folding run side-by-side on the same job on Pro.
Every Δ near the ±1 kcal/mol Vina noise floor gets a within-noise badge; mutations outside the pocket are flagged, not scored; every pose comes with a plain-English readout and an inline ADMET panel (hERG, DILI, CYP, BBB). We publish our method limits on purpose.
Docking is fast triage; MM/GBSA is the next rung of confidence. We take the docked pose and do a single-snapshot rescore with implicit-solvent molecular mechanics (Amber ff14SB / OpenFF Sage 2.2 / OBC2, εin = 1, no salt), then rank your compounds by the binding-energy estimate. A separate short MD run checks the pose actually holds — reported as a stability badge, never averaged into the score.
Validated across 8 congeneric series (184 ligands): median Spearman ρ 0.71, Pearson r 0.67, ranging from HIF2A ρ 0.46 to TYK2 ρ 0.83 — rescoring reference poses. We lead with rank correlation because that is the honest measure of a rescoring signal, and we publish every target, including the weak ones.
Each dot is one ligand; the dashed line is the least-squares fit. More-negative = stronger binding. Curated reference poses.
Case study (TYK2, n = 12): our template-anchored docking — which uses the co-crystal ligand as a reference but docks the query itself rather than borrowing a crystal pose — followed by MM/GBSA rescoring reaches ρ 0.72, versus ρ 0.83 when the same ligands are rescored from reference poses. One series so far; other targets still run the de-novo docking path.
MM/GBSA magnitudes run several-fold larger than experimental ΔG and omit configurational entropy, so read this as a way to order compounds within a target, not an absolute affinity. For the close calls, FEP is available by request.
MM/GBSA scoring is available by request.
| Free servers | Liganx | Schrödinger Maestro | |
|---|---|---|---|
| Compound resistance forecast across a variant panel | — | ✓ | partial |
| Runs in the browser, no install | ✓ | ✓ | — |
| Mutation-aware WT-vs-mutant matrix | — | ✓ | partial |
| Public resistance-mutation atlas | — | ✓ | — |
| Pre-computed FDA-drug screenings | — | ✓ | — |
| Bulk virtual screening, selectivity-ranked | — | ✓ | partial |
| Inline ADMET (hERG / DILI / CYP / BBB) | — | ✓ | partial |
| Multiple scoring engines side-by-side | — | ✓ | partial |
| Ensemble / flexible-receptor docking | — | ✓ | ✓ |
| Published, reproducible validation report | — | ✓ | — |
Reflects publicly known features as of May 2026. Free-server and Schrödinger capabilities vary by version, license tier, and module.
One UI. Real Vina under the hood. Selectivity matrix in minutes, not days.