Similarity Searching - Usage Guide

Overview

Find structurally similar molecules using fingerprint-based Tanimoto similarity. Includes library searching, clustering, and maximum common substructure analysis.

Prerequisites

pip install rdkit
pip install numpy pandas

Quick Start

Tell your AI agent what you want to do: - "Find compounds similar to my lead molecule" - "Search my library for molecules with Tanimoto > 0.7 to this query" - "Cluster my compound library by structural similarity" - "Find the maximum common substructure among these molecules"

Example Prompts

Similarity Search

"Find all compounds in my library with Tanimoto similarity > 0.7 to aspirin."

"Search for analogs of this SMILES using ECFP4 fingerprints."

Clustering

"Cluster my compound library at 70% similarity using Butina clustering."

"Group similar molecules together with a 0.3 distance cutoff."

Scaffold Analysis

"Find the maximum common substructure for my hit series."

"What structural features are shared by these active compounds?"

What the Agent Will Do

1. Generate fingerprints for query and library
2. Calculate pairwise Tanimoto similarities
3. Filter/rank by similarity threshold
4. Perform clustering if requested
5. Find MCS for shared scaffold analysis

Tips

• For ECFP4, >0.85 and >0.7 are repository starting heuristics for close and related analogs; calibrate them on the project's known series and fingerprint settings
• Butina distance cutoff = 1 - centroid-neighbor similarity threshold (cutoff 0.3 assigns neighbors at >=0.7 similarity to a selected centroid; it does not guarantee >=0.7 for every member pair)
• BulkTanimotoSimilarity is faster for large libraries
• ECFP4 (radius=2) is the most common fingerprint for similarity
• MCS timeout should be set for large molecule sets

Related Skills

• molecular-descriptors - Generate fingerprints for similarity
• substructure-search - Pattern-based searching
• molecular-io - Load molecules for searching