Predict HDAC inhibitor potency & selectivity from structure alone.
Paste a SMILES string. Get physicochemical descriptors, drug-likeness flags, and estimated binding potency across four HDAC isoforms — from models trained on real bioactivity data.
Select targets to predict
Enter a SMILES string
Predicted potency by isoform
Batch mode — upload a CSV
CSV needs a smiles column (or SMILES in the first column). Up to 30 rows (structure images make each row slower to render).
Why HDAC11
Zinc-dependent enzymes, and one understudied isoform.
Histone deacetylases remove acetyl groups from lysine residues on histone tails, tightening chromatin and switching genes off. Their catalytic core coordinates a single zinc ion — the same site most inhibitors, including hydroxamic acids, are designed to chelate.
HDAC1, HDAC6, and HDAC8 are relatively well-studied, with thousands of known inhibitors in ChEMBL. HDAC11 remains comparatively underexplored, despite emerging interest in immuno-oncology. Comparing predicted potency across all four in one place — a rough isoform selectivity estimate — is something almost no free public tool offers; most HDAC QSAR work exists only inside published papers, not as something a chemist can actually open and use.
How it works
Structure in, estimated potency out.
The model is trained on public HDAC11 bioactivity data from ChEMBL — real IC50 and Ki measurements from published assays, not synthetic or simulated data.
Structure fingerprint
RDKit converts the SMILES into descriptors and a Morgan fingerprint — a structural signature the model can read.
Trained prediction
A separate model per isoform — trained on 500 to 8,000+ known compounds per target, depending on how well-studied it is — estimates binding potency from that fingerprint.
Confidence check
Each prediction is scored by similarity to the training data — so you know when it's extrapolating.
On accuracy — models are trained per isoform on ChEMBL data (HDAC1: 8,282 compounds, test R² 0.67 · HDAC6: 6,515 compounds, R² 0.66 · HDAC8: 3,097 compounds, R² 0.63 · HDAC11: 524 compounds, R² 0.49). Treat predictions as a rough ranking, not a lab measurement — selectivity comparisons are estimates, not confirmed selectivity data. Descriptors and drug-likeness flags are exact RDKit calculations, independent of any model.