Chemistry and compounds · What you can do here
Screen a compound library against a target
Screen compound libraries against protein targets using AutoDock Vina, RDKit, and enrichment validation without local installation.
What this involves
Virtual screening evaluates how chemical libraries fit a protein binding pocket, ranking candidates for synthesis or biochemical assay. The process requires careful cheminformatics triage, structural receptor preparation, conformational search, and pose analysis. Docking addresses where and how a small molecule binds rather than yielding true binding free energy.
Doing this properly requires curating both target and library before running search calculations. Compounds must be desalted, canonicalised, protonated, and filtered for property profiles or triage alerts. The receptor structure requires verified resolution, preserved structural cofactors and bridging waters, and specific protonation states. Blind runs or unvalidated boxes compromise results.
Crucially, the ranking capacity of the screen must be validated before interpreting hits. Redocking the co-crystallised ligand confirms that the protocol reproduces known crystallographic poses, while decoy benchmarking verifies that active molecules enrich against property-matched background compounds.
How it goes
- 1Receptor Selection and PreparationRetrieve experimental structures from the PDB with resolution ≤ 2.5 Å, retaining crucial cofactors, structural waters, and correct tautomer states at pH 7.4.
- 2Library Standardisation and Cheminformatics TriageDesalt, uncharge, and standardise compounds with RDKit, filtering by Lipinski Ro5, Veber rules, or PAINS alerts and generating initial 3D conformers via ETKDGv3.
- 3Search Box Definition and Control RedockingDefine the docking grid around the co-crystallised ligand centroid and redock the cognate ligand, requiring symmetry-corrected RMSD < 2.0 Å using CalcRMS without superposition.
- 4Enrichment BenchmarkingValidate screening capacity by docking known actives against property-matched decoys, verifying performance with EF1% and BEDROC(α=20) metrics.
- 5Virtual Screening ExecutionDock library conformers using AutoDock Vina, SMINA, or DiffDock across three independent random seeds with exhaustiveness set to ensure convergence.
- 6Pose Plausibility and Ligand Efficiency ScoringRank candidates by ligand efficiency and docking score, discarding poses with high internal MMFF strain, buried uncompensated charges, or bad contacts.
Start it here
Ask in plain language. The agent works on a Linux machine of its own — it installs and runs what the job needs, keeps every file, and reads the results back with the numbers explained. Nothing to set up, and a long run continues after you close the page.
What to watch for
- Never convert a docking score into Kd, Ki, or IC50 values using RT ln K or rough conversions, as scores do not equal binding free energies.
- Stripping catalytic metals, required cofactors, or conserved bridging waters from the binding site will produce confident nonsense.
- Evaluating redocking poses with superposition algorithms like GetBestRMS will mask failures by reporting artificially low RMSD values for wrong-pocket placements.
- A docking run where top scores vary by more than 1 kcal/mol across independent random seeds has failed to converge and requires higher exhaustiveness.
- Docked conformers displaying internal MMFF strain exceeding 10 kcal/mol relative to their locally minimised states indicate spurious, unphysical poses.
- Filtering compound libraries silently without documenting which rules or PAINS alerts triggered exclusions hides useful borderline molecules.
What it runs
Models
- AutoDock VinaApache-2.0Predicts small-molecule binding poses and affinity scores against protein structures.
- SMINAApache-2.0A molecular docking tool optimised for fragment-sized ligands and custom scoring functions.
- DiffDockMITDiffusion-based generative model predicting 3D poses of small-molecule ligands bound to proteins.
- MolFormerApache-2.0A chemical language transformer that extracts molecular embeddings from SMILES strings for property prediction.
Databases it can query
Questions
Do I need to install anything to screen a compound library against a target?
No. The research agent does this work on a Linux machine of its own — it installs what the job needs, keeps the files, and reads the results back to you. Long runs continue if you close the page.
Can docking scores predict binding affinities or rank compounds across different targets?
No. Vina-family scores correlate weakly with affinity (r ≈ 0.4–0.5) with typical errors of 2–3 kcal/mol, and scores scale with molecular size. Docking scores must only be compared within the same target, search box, and engine, ideally reported alongside ligand efficiency.
How are active and decoy libraries prepared for validation?
Known actives identified via databases such as ChEMBL or BindingDB are paired with approximately 50 decoys per active, matched on molecular weight, logP, charge, and rotatable bonds while remaining topologically dissimilar.
What threshold determines whether a redocked positive control has succeeded?
A symmetry-corrected RMSD below 2.0 Å without superposition indicates the crystallographic pose was successfully reproduced.
Can covalent binders or macrocycles be screened using this standard docking protocol?
No. Standard protocols do not support covalent inhibitors, metal-chelating warheads, or macrocycles, as rings are kept rigid during docking search routines and specialized parameterisation is required.
Written from the method the research agent follows for this work and the documentation of the tools listed above. Statements citing a figure that does not appear there are removed before publication. Corollary is a non-commercial research demonstration; results are not for clinical use.