Chem Similarity Search

Chemical Similarity Search

Goal

To programmatically find chemical analogs, alternative precursors, and structurally similar compounds for a given target molecule using PubChem's "fastsimilarity_2d" endpoint. The skill retrieves lists of similar compounds ranked by sequence alignment of their 2D molecular fingerprints, providing CIDs, molecular weights, formulas, and SMILES strings.

Instructions

1. Search by SMILES String

Search for similar compounds by providing the canonical or isomeric SMILES. Adjust the --threshold (similarity cutoff 0-100, default is 95) to widen or narrow the search radius. Higher threshold equals higher similarity. Adjust --max_records to limit the output length.

# Env: base-agent
python .agents/skills/chem-similarity-search/scripts/similarity_search.py \
  --smiles "CC(=O)Oc1ccccc1C(=O)O" \
  --threshold 95 \
  --max_records 5 \
  --outdir research/aspirin_similar \
  --output aspirin_similar.json

2. Search by PubChem CID

Search directly using an exact compound's CID. This avoids translation steps for SMILES parsing.

# Env: base-agent
python .agents/skills/chem-similarity-search/scripts/similarity_search.py \
  --cid 2244 \
  --threshold 90 \
  --max_records 10 \
  --outdir research/aspirin_similar \
  --output cid_2244_similar.json

Examples

We can test extracting highly similar analogs (Threshold 95) for Aspirin (CID: 2244 or SMILES: CC(=O)Oc1ccccc1C(=O)O).

# Env: base-agent
python .agents/skills/chem-similarity-search/scripts/similarity_search.py \
  --cid 2244 \
  --threshold 95 \
  --max_records 5 \
  --outdir .agents/skills/chem-similarity-search/examples/aspirin_analogs \
  --output aspirin_analogs.json

Constraints

• Rate Limiting: PubChem PUG REST API enforces per-user throttling limits. Heavy bursts will result in HTTP 503 Server Busy errors. The script implements an exponential backoff retry mechanism.
• 2D Similarity: Uses exact structural bit-vector fingerprints. Stereochemical and 3D properties do not strongly affect the score.
• Network: Internet access is required.

---

---

Author: Bowen Deng Contact: GitHub @learningmatter-mit

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.