Drug Molecular Fingerprints

Molecular Fingerprints

Goal

To compute circular Morgan fingerprints (ECFP-style; default ECFP4 with radius=2) for a set of compounds, then calculate pairwise Tanimoto similarity for library comparison. Optionally perform Butina clustering for diversity analysis and generate a similarity heatmap for small sets.

This skill is commonly used for hit expansion, SAR triage, compound library diversity assessment, and applicability-domain style analysis.

Instructions

The drugdisc MCP server provides a compute_molecular_fingerprints tool that can be called directly:

Basic usage with SMILES file:

mcp_drugdisc_compute_molecular_fingerprints(
    smiles_file="compounds.smi",
    radius=2,
    fp_size=2048,
    compute_similarity=True,
    output_file="similarity.json"
)

With Butina clustering:

mcp_drugdisc_compute_molecular_fingerprints(
    smiles_file="library.smi",
    cluster=True,
    cluster_cutoff=0.7,
    output_file="clustered.json"
)

With similarity heatmap (small molecule sets, ≤250 compounds):

mcp_drugdisc_compute_molecular_fingerprints(
    smiles_file="hits.smi",
    save_heatmap="heatmap.png",
    output_file="similarity.json"
)

Feature Morgan (FCFP-like) fingerprints:

mcp_drugdisc_compute_molecular_fingerprints(
    smiles_file="compounds.smi",
    use_features=True,
    output_file="fcfp_similarity.json"
)

Chirality-aware fingerprints:

mcp_drugdisc_compute_molecular_fingerprints(
    smiles_file="enantiomers.smi",
    use_chirality=True,
    output_file="chiral_sim.json"
)

Examples

SMILES file format

CCO	ethanol
CCCO	propanol
c1ccccc1	benzene
c1ccc(cc1)O	phenol

Basic similarity analysis

mcp_drugdisc_compute_molecular_fingerprints(
    smiles_file=".agents/skills/drug-molecular-fingerprints/examples/compounds.smi",
    output_file="similarity.json"
)

Diversity-based clustering for library selection

mcp_drugdisc_compute_molecular_fingerprints(
    smiles_file="screening_library.smi",
    cluster=True,
    cluster_cutoff=0.5,
    output_file="diverse_clusters.json"
)

Output Format

The tool returns a JSON with:

• n_compounds: Total number of input compounds
• n_valid: Number of successfully processed compounds
• compounds: List of compound info (SMILES, name, validity, fingerprint bits)
• similarity_matrix: Pairwise Tanimoto similarity (if compute_similarity=True)
• clusters: Butina clustering results (if cluster=True)

Constraints

• MCP Server: Requires drugdisc MCP server
• Dependencies: RDKit (Chem, rdFingerprintGenerator, DataStructs, ML.Cluster.Butina)
• SMILES file format: One molecule per line, SMILES[whitespace]NAME (NAME optional), # for comments
• Fingerprint defaults: Morgan radius=2 (ECFP4-like), 2048 bits
• Heatmap rendering: Limited to ≤250 compounds due to memory constraints
• Similarity metric: Tanimoto coefficient (Jaccard index for bit vectors)
• Clustering algorithm: Butina (leader-picker style); cutoff = similarity threshold (not distance)

---

Author: Matthew Cox Contact: GitHub @mcox3406

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: learningmatter-mit
• Source: learningmatter-mit/AtomisticSkills
• License: MIT
• Homepage: https://arxiv.org/abs/2605.24002

Install and usage instructions live in the source repository linked above.