| name | rdkit |
| description | Cheminformatics toolkit for fine-grained molecular control. Parse and write SMILES, SDF, MOL and InChI; compute descriptors (MW, LogP, TPSA, QED, Bertz); build fingerprints (Morgan/ECFP, RDKit, MACCS, atom pair, torsion) and score Tanimoto, Dice or cosine similarity; run SMARTS substructure search and reaction SMARTS; generate 2D depictions and ETKDG 3D conformers; extract Murcko scaffolds and canonical hashes; control sanitization and stereochemistry directly. Also trigger on rdkit, Chem.MolFromSmiles, rdFingerprintGenerator, SDMolSupplier, SMARTS query, ETKDG, or FilterCatalog. For standard workflows with a simpler interface use the datamol skill, which wraps RDKit; use rdkit for advanced control, custom sanitization, and specialized algorithms. |
| license | MIT |
| allowed-tools | Read Write Edit Bash |
| compatibility | Examples target RDKit 2026.03.x. Use conda-forge for the broadest binary support or PyPI package `rdkit` for supported platform wheels; `rdkit-pypi` is the legacy PyPI name. |
| metadata | {"version":"1.4","skill-author":"K-Dense Inc."} |
RDKit Cheminformatics Toolkit
Overview
RDKit is a comprehensive cheminformatics library providing Python APIs for molecular analysis and manipulation. This skill provides guidance for reading/writing molecular structures, calculating descriptors, fingerprinting, substructure searching, chemical reactions, 2D/3D coordinate generation, and molecular visualization. Use this skill for drug discovery, computational chemistry, and cheminformatics research tasks.
Checked against: RDKit 2026.03.5 (rdkit 2026.3.5 on PyPI, released 2026-08-03), August 2026. Official installation docs continue to recommend conda-forge for most users, while cross-platform PyPI wheels are published under the rdkit package name. rdkit-pypi is the old PyPI package name and should only appear when maintaining legacy environments.
Installation and Setup
Use uv when installing into an existing Python environment:
uv pip install rdkit
For reproducible chemistry environments, especially when mixing compiled scientific packages, conda-forge remains the upstream recommendation:
conda create -c conda-forge -n my-rdkit-env rdkit
conda activate my-rdkit-env
Avoid installing both conda rdkit and PyPI rdkit/rdkit-pypi into the same environment unless you are deliberately debugging packaging behavior. Mixed installs can make it unclear which binary extension is being imported.
Core Capabilities
Twelve capability areas, each with worked code, are documented in
references/core_capabilities.md:
| # | Area | Covers |
|---|
| 1 | Molecular I/O and creation | SMILES, MOL files and blocks, InChI, SDF and SMILES suppliers, multithreaded reading, writers |
| 2 | Sanitization and validation | disabling automatic sanitization, manual and partial sanitization, detecting problems first |
| 3 | Analysis and properties | atom and bond iteration, ring information and SSSR, chirality and stereochemistry, fragments |
| 4 | Descriptors | MW, LogP, TPSA, H-bond donors/acceptors, rotatable bonds, aromatic rings, bulk calculation, drug-likeness |
| 5 | Fingerprints and similarity | topological, Morgan/ECFP via rdFingerprintGenerator, MACCS, atom pair, torsion, Avalon; Tanimoto and other metrics; Butina clustering |
| 6 | Substructure searching | SMARTS queries, match retrieval, and a library of common patterns |
| 7 | Chemical reactions | reaction SMARTS, applying reactions, reaction fingerprints |
| 8 | 2D and 3D coordinates | depiction, template alignment, ETKDG embedding, force-field optimization, RMSD, constrained embedding |
| 9 | Visualization | single and grid images, substructure highlighting, custom drawer options, Jupyter integration, fingerprint bit environments |
| 10 | Molecular modification | explicit hydrogens, Kekulization, aromaticity, substructure replacement, charge neutralization |
| 11 | Hashes and standardization | Murcko scaffold and canonical hashes, regioisomer hashes, randomized SMILES for augmentation |
| 12 | Pharmacophore and 3D features | feature factories and feature extraction |
Worked workflows and the performance, thread-safety, and version-sensitivity notes are in
references/workflows_and_best_practices.md.
Prefer portable exchange formats (SMILES, SDF) for shared data; for local caches RDKit's
binary molecule representation avoids generic pickle.
Common Pitfalls
- Forgetting to check for None: Always validate molecules after parsing
- Sanitization failures: Use
DetectChemistryProblems() to debug
- Missing hydrogens: Use
AddHs() when calculating properties that depend on hydrogen
- 2D vs 3D: Generate appropriate coordinates before visualization or 3D analysis
- SMARTS matching rules: Remember that unspecified properties match anything
- Thread safety with MolSuppliers: Don't share supplier objects across threads
Resources
references/
All five bundled reference documents, loaded only when needed:
Only the files listed in references/ and scripts/ are bundled local resources. Names such as rdkit, datamol, scipy, and sklearn refer to installable Python packages, not local files in this skill.
scripts/
python skills/rdkit/scripts/molecular_properties.py "CC(=O)Oc1ccccc1C(=O)O"
python skills/rdkit/scripts/molecular_properties.py --file library.smi --output properties.csv
python skills/rdkit/scripts/similarity_search.py "c1ccccc1O" library.sdf --threshold 0.6
python skills/rdkit/scripts/substructure_filter.py library.smi --pattern "C(=O)[OH]" --report hits.csv
Each exits nonzero and writes to stderr on failure, so they compose in a pipeline. They are equally
usable as templates for custom workflows.
Composing with the rest of the bundle
datamol → instead: the same standardization, clustering and parallel work with sensible
defaults. Reach for rdkit only when you need control datamol does not expose.
medchem → after: real triage. Its rule catalogue and the full PAINS/NIBR alert sets are what you
want for library filtering; substructure_filter.py here is a general SMARTS tool, not a
curated alert set.
molfeat → after: turning molecules into model-ready features rather than hand-rolled fingerprints.
chembl → before: measured bioactivity to featurize, rather than a library you invented.
chemical-space → after: once a SMARTS query defines the chemotype, find purchasable examples.
admet-prediction / deepchem / pytdc → after: descriptors and fingerprints from here are the
input those models expect. Desalt and standardise first or you predict on the wrong species.