Guide · Agent skill
How to use the Datamol cheminformatics agent skill
Updated · written with AI from the project's own documentation and checked against our catalogue data
The datamol agent skill teaches an AI agent to work with chemical structures written as text, such as SMILES strings, using the datamol Python library, a simplified wrapper around RDKit. It is aimed at medicinal chemists, cheminformatics students and data scientists who want sensible-default scripts for molecule datasets. After this guide you will know what the skill does, how to install it, how to phrase requests so the agent uses it, and what to check before trusting the output.
Datamol Cheminformatics ToolkitK-Dense-AI/scientific-agent-skills · MIT Download the skill Quick start
- 1Download and unzip the skill. Download the skill ZIP from this page and unzip it into ~/.claude/skills/ or a project's .claude/skills/ folder.
- 2Or upload it in the Claude apps. In the Claude apps, go to Settings → Capabilities → Skills and upload the skill.
- 3Install datamol. Run uv pip install "datamol==0.13.0" in an isolated Python 3.11+ environment.
- 4Ask in plain language. Describe the chemistry task, the input (for example SMILES in a CSV) and the output you want.
- 5Check parse results. Ask the agent to report invalid SMILES and keep source IDs with each result.
- 6Review the code before running it. Read the generated Python, especially any remote paths and standardization steps, before running it.
What is an agent skill and what does this one do?
An agent skill is a folder of instructions and reference files that an AI agent loads when a task matches it. This skill is the datamol skill from the K-Dense-AI scientific-agent-skills collection, published under the MIT licence. Once installed, the agent knows how to write Python code that uses datamol for molecule work.
The skill covers molecular format conversion (SMILES, SELFIES, InChI), structure standardization, descriptors and fingerprints, clustering and diversity selection, scaffold and fragment analysis, 3D conformer generation, drawing structures, applying reaction rules, and reading and writing SDF, CSV and Excel files. Every molecule it returns is a native rdkit.Chem.Mol object.
Install the datamol skill (ZIP download or Claude apps)
Download the skill ZIP from this page and unzip it. For a personal setup, place the unzipped folder in ~/.claude/skills/. For a single project, place it in that project's .claude/skills/ folder. In the Claude apps, you can instead upload the skill under Settings → Capabilities → Skills.
The skill itself only holds instructions. Running the code it writes needs Python 3.11 or newer and the datamol library, which the skill tells the agent to install with uv pip.
- Personal: ~/.claude/skills/
- Project: .claude/skills/ inside the project folder
- Claude apps: Settings → Capabilities → Skills, then upload
Install the datamol library the skill uses
The skill's setup instructions use this command. Installing needs network access. Use an isolated environment, because scientific packages can have conflicting requirements.
RDKit is installed as a dependency. The skill requires RDKit 2024.09 or newer.
uv pip install "datamol==0.13.0"How do you trigger the skill with a natural request?
You do not need special syntax. Describe the chemistry task in plain language and mention molecules as SMILES where you have them. The agent matches the request to the skill's description and loads it.
Good requests name the input, the operation and the output file or table you want. Mentioning datamol directly makes the match more reliable.
- Standardize these SMILES in a CSV and report which rows fail to parse.
- Compute the standard descriptors and ECFP4 fingerprints for compounds.csv and save to Excel.
- Pick 50 diverse compounds from this SDF using datamol.
- Group these molecules by Bemis-Murcko scaffold and count each group.
- Generate 10 3D conformers for this molecule and compute SASA.
What examples does the skill cover?
The skill documents ten workflow areas with worked code, including basic molecule handling, file reading and writing, descriptors, fingerprints and similarity, clustering and diversity, scaffold analysis, fragmentation, 3D conformers, visualization and chemical reactions. Three end-to-end pipelines cover load, filter and analyse; SAR by scaffold series; and virtual screening.
For detailed function signatures, the agent consults the reference files shipped with the skill, such as core_api.md, io_module.md and conformers_module.md.
- Basic handling: to_mol, batch conversion and canonical SMILES
- Files: SDF, SMILES, CSV and Excel, including s3:// and GCS paths
- Similarity: ECFP4 fingerprints and nearest-neighbour lookup
- Scaffolds: scaffold-disjoint train/test splits for machine learning
What does the skill need, and what are the safety notes?
Local molecular workflows need no credentials. Remote reading or writing, for example from S3 or GCS, needs the provider credentials you choose to use. The skill runs scripts and uses the network during installation.
Standardization can change the entity you are measuring. Metal disconnection, salt stripping, neutralization and stereochemistry changes are not applied automatically for organometallic or formulation work. Keep original structures and IDs next to transformed ones. The skill performs computational chemistry only; it does not give validated safety or regulatory conclusions.
- Python 3.11 or newer, datamol 0.13.0 and RDKit 2024.09 or newer
- Network access for installation
- Provider credentials only for remote paths
- Keep source IDs and a rejection log for invalid SMILES
Tips for reliable results
Pin the environment. RDKit upgrades can change canonical SMILES and retained conformers, so compare chemical invariants rather than exact strings or conformer counts.
For large batch descriptor work, the skill notes that a positive batch_size (for example 128) is needed when parallelism is enabled. For very large datasets, Butina clustering needs memory that grows with the square of the number of molecules, so use pick_diverse with a bounded number of picks instead.
- Always check for None after parsing before downstream steps
- Use n_jobs=-1 for all cores, or n_jobs=1 for sequential runs
- For advanced control or custom parameters, ask for raw RDKit instead
What people use it for
- Standardize and deduplicate a CSV of screening compounds, logging rows that fail to parse.
- Compute standard descriptors and ECFP4 fingerprints in parallel for a machine learning dataset.
- Pick a diverse subset of a large SDF library for a follow-up screen.
- Split a compound set into scaffold-disjoint train and test sets.
- Generate and minimize 3D conformers for a molecule and compute its solvent-accessible surface area.
Frequently asked questions
How do I install the datamol agent skill?
Download the skill ZIP from this page and unzip it into ~/.claude/skills/ for personal use, or into a project's .claude/skills/ folder. In the Claude apps you can instead upload it under Settings → Capabilities → Skills.
What do I need to run datamol?
The skill requires Python 3.11 or newer and datamol 0.13.0, with RDKit 2024.09 or newer installed as a dependency. Installation needs network access. Local workflows need no credentials.
Is the datamol skill free?
The skill is published under the MIT licence, as listed on this page. The source does not state pricing for anything else, so check the licence file you download for terms.
Does datamol replace RDKit?
No. Datamol is a wrapper around RDKit with simpler defaults and returns native rdkit.Chem.Mol objects. For advanced control or custom parameters, the skill says to use RDKit directly.
Why does the agent say a SMILES is invalid?
to_mol returns None when parsing fails. The skill recommends checking for None after every parse and keeping the source row in a rejection log, so you can inspect the original input.
Should I use this for organometallic or formulation work?
The skill advises against it. Automatic metal disconnection, salt stripping or neutralization can change the entity being assayed, so those transformations should not be applied automatically in those cases.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
Use this skill if you want an agent to write quick, sensible-default datamol scripts for standard small-molecule datasets, such as cleaning, descriptors, fingerprints, clustering and scaffold splits. Pick raw RDKit if you need custom parameters, and avoid relying on it for organometallic or formulation chemistry or for safety and regulatory conclusions.
Download the skill