synthetic-sciences/openscience/smiles-validation

SMILES Validation

Overview

LLMs frequently generate invalid SMILES or produce molecules that don't match their stated reasoning. This skill provides strict validation, structural comparison, and modification verification.

Key checks:

• Parse validation: RDKit sanitization, valence checking, parenthesis/bracket balance
• Structural comparison: Tanimoto similarity, scaffold preservation, MCS analysis
• Modification verification: Confirm claimed structural changes exist in the actual molecule
• Classification: Categorize changes as optimization (>0.6 similarity), significant modification (0.4-0.6), or de novo design (<0.4)

When to Use This Skill

• After molecule generation: Validate every SMILES before reporting results
• Optimization verification: Confirm proposed modifications match the actual structure
• Batch validation: Check a library of generated molecules for validity
• Quality control: Ensure reproducibility of molecular designs

Installation

pip install rdkit-pypi

Core Workflows

1. Validate a Single SMILES

python scripts/validate.py --smiles "c1ccccc1"

2. Compare Original vs Modified

python scripts/validate.py --original "c1ccccc1" --proposed "c1ccc(O)cc1" --check-modification "Added hydroxyl group"

3. Batch Validation

python scripts/validate.py --input generated_molecules.csv --output validation_report.json

Script Reference