RDKit

What it does

RDKit is an open-source cheminformatics library, primarily used via Python, that provides the core operations for working with molecular structures computationally. If Chemprop, IBM RXN, or REINVENT 4 are the destination, RDKit is the road: most chemistry AI tools use RDKit under the hood for reading, writing, and processing molecular structures.

Core capabilities:

• Molecule I/O: read and write SMILES, InChI, SDF, MOL2, and other standard formats
• Structure manipulation: add/remove atoms and bonds, compute ring systems, enumerate stereoisomers, generate 2D and 3D conformers
• Molecular descriptors: compute hundreds of physicochemical properties (molecular weight, logP, TPSA, rotatable bonds, hydrogen bond donors/acceptors)
• Fingerprints: generate Morgan (circular), MACCS keys, RDKit, and other fingerprints for similarity search and ML feature generation
• Substructure search: find molecules containing a specified substructure using SMARTS patterns
• Similarity and clustering: compute Tanimoto similarity between molecules, cluster compound sets
• Reaction handling: parse, manipulate, and apply chemical reactions

Best for

Computational chemists and cheminformaticians who need to manipulate and analyze molecular datasets in Python. RDKit is the standard prerequisite for most Python-based cheminformatics work — if you’re building a chemistry ML pipeline, processing compound libraries, or preparing data for Chemprop or similar tools, you’ll use RDKit.

Pricing

Free and open-source (BSD 3-Clause license). Available via conda (conda install -c conda-forge rdkit) or pip.

Strengths

• De facto standard: virtually all Python-based cheminformatics and chemistry ML tools depend on or interface with RDKit; learning it unlocks access to the entire ecosystem
• Comprehensive: covers the full range of cheminformatics operations from molecule parsing to 3D conformer generation without needing additional tools for standard tasks
• Fast: core operations are implemented in C++ with Python bindings — suitable for processing millions of compounds
• Active development and large community: well-documented, regular releases, extensive tutorials and notebooks available
• Integrates with pandas, NumPy, scikit-learn, PyTorch — fits naturally into Python data science workflows

Limitations

• It’s a library, not a tool: RDKit requires programming — there is no graphical interface; researchers without Python experience will find it inaccessible without support
• Not designed for AI/ML out of the box: RDKit computes features and fingerprints, but training models on those features requires scikit-learn, PyTorch, or similar; Chemprop and similar tools handle the ML layer
• 3D conformer generation is functional but not as accurate as specialized tools (OpenEye OMEGA, Schrodinger ConfGen) for high-quality conformer ensembles
• Documentation assumes cheminformatics familiarity; the learning curve for researchers from wet-lab backgrounds can be steep

How it compares

Related content

• Field Guide: Chemistry
• Tool: Chemprop — ML property prediction built on molecular graph representations
• Tool: IBM RXN — uses reaction SMILES that RDKit can generate and parse
• Glossary: SMILES Notation
• Glossary: Molecular Fingerprint