Screening methods
2D similarity & substructure
Fingerprint & Substructure Search
Rank every database molecule by its structural similarity to one or more known actives, or retrieve all molecules that contain a scaffold or functional group.
- Morgan (ECFP4) fingerprints, 2,048 bits; Tanimoto similarity
- Substructure queries as SMILES or SMARTS, optional tautomer enumeration
- Hit counts at every Tanimoto cut-off; 3D SDF extraction with a drug-like filter
- Excel hit lists, SDF files and PDF reports
A cathepsin S inhibitor and DrugBank compounds found by similarity (Tanimoto 0.56–0.67).
Search results: the hits per database and Tanimoto cut-off, ready for SDF extraction.
3D
Pharmacophore Screening
Describe what a ligand needs to bind — hydrogen-bond donors and acceptors, hydrophobic and aromatic groups, charges, in their 3D arrangement — and find molecules of any scaffold that can match it.
- Models from a protein–ligand complex (e.g. a PDB entry), a docked pose or active ligands; editable feature by feature
- 3D conformer ensembles (CONFORGE) for your libraries; pharmacophore-ready databases of natural products, drugs and metabolites
- Model validation with actives and decoys: ROC, AUC, enrichment factors
- Aligned hit poses, sent directly to docking or QSAR prediction
Built on CDPKit (University of Vienna).
A pharmacophore from the cathepsin S crystal structure 2FQ9: eight features on the bound inhibitor, each editable.
Matching: a hit places its features inside the model's tolerance spheres.
Ranked hits with their aligned 3D poses.
Machine learning
QSAR Modelling & Prediction
Learn the relation between structure and measured activity for your target, and predict the activity of new molecules — with the validation reviewers expect.
- Curated datasets from ChEMBL or your own data (standardisation, duplicates, units)
- RDKit, Mordred and PaDEL descriptors, fingerprints
- MLR, PLS, random forest, extra trees, gradient boosting, SVR, kNN; classification models
- Scaffold / random splits, double cross-validation, Y-randomisation, Golbraikh–Tropsha criteria, applicability domain (Williams plot)
- Predictions for SMILES, files, libraries and pharmacophore hits; OECD-style PDF report
Model comparison for 2,437 cathepsin S inhibitors: observed vs predicted, Williams plot, Y-randomisation, descriptor importance.
Test set set aside first; double cross-validation on the rest.
Candidates with predicted activity and domain flags.
Triage
Drug-likeness & ADMET Screening
Shortlist hits by the properties that decide whether a molecule can become a drug.
- Lipinski, Veber, Ghose and QED drug-likeness
- Physicochemical properties: MW, logP, TPSA, H-bond donors / acceptors, rotatable bonds
- Structural alerts (PAINS, Brenk)
- Interactive tables, 3D viewer, PDF reports; selected molecules sent to docking
Drug-like screening of a hit set with pass / fail per rule.
ADMET & drug-likeness of individual molecules.