Claude Code Plugins

Community-maintained marketplace

Feedback

|

Install Skill

1Download skill
2Enable skills in Claude

Open claude.ai/settings/capabilities and find the "Skills" section

3Upload to Claude

Click "Upload skill" and select the downloaded ZIP file

Note: Please verify skill by going through its instructions before using it.

SKILL.md

name structure-search
description Structure-based similarity search and scaffold analysis for drug discovery. Use for lead hopping, scaffold morphing, and chemical space exploration. Keywords: similarity search, scaffold hopping, chemical space, fingerprint, Tanimoto
category Computational Chemistry
tags structure, similarity, fingerprint, scaffold, chemical-space
version 1.0.0
author Drug Discovery Team
dependencies rdkit, chembl, pubchem

Structure Search Skill

Structure-based similarity search and scaffold analysis for drug discovery.

Quick Start

/structure --query "CC1=CC=C(C=C1)CNC" --threshold 0.7
/scaffold-hop --input compound.sdf --scaffold-type murcko
/similar-compounds --chembl "CHEMBL210" --limit 20

Similarity Methods

Fingerprint-Based

Fingerprint Size Best Use Speed
Morgan 2048 General purpose Fast
MACCS 166 General purpose Very fast
RDKit 2048 Structural features Fast
Atom pair 2048 Substructure Medium
Topological torsion 2048 Conformations Medium

Similarity Coefficients

Coefficient Range Properties
Tanimoto 0-1 Most common, bounded
Dice 0-1 Similar to Tanimoto
Cosine 0-1 Vector-based
Tversky 0-1 Asymmetric

Scaffold Analysis

Scaffold Types

Type Definition Use Case
Murcko Core ring system General
Bemis-Murcko Rings + linkers Drug-like
RECAP Rings + functional groups Medicinal chemistry
Graph Only topology Very generic

Scaffold Hopping

Strategies:

  1. Ring replacement: Bioisosteric substitution
  2. Ring opening/closing: Modify topology
  3. Linker modification: Change connectivity
  4. Heteroatom swap: N→O→S→C

Output Structure

# Structure Search Results

## Query Compound
**SMILES**: CC1=CC=C(C=C1)CNC
**Name**: Erlotinib
**Scaffold**: c1ccc(cc1)CNCC

## Similar Compounds (Tanimoto ≥ 0.7)

| Rank | ID | Name | Similarity | Scaffold Match |
|------|----|-----|------------|----------------|
| 1 | CHEMBL210 | Erlotinib | 1.00 | Yes |
| 2 | CHEMBL214 | Gefitinib | 0.89 | Yes |
| 3 | CHEMBL617 | Afatinib | 0.82 | Yes |
| 4 | CHEMBL12345 | Novel analog | 0.76 | No |
| 5 | CHEMBL98765 | Lead compound | 0.72 | Yes |

## Scaffold Analysis

### Murcko Scaffold

Query: c1ccc(cc1)CNCC (Quinazoline core)


### Known Compounds with This Scaffold

| Compound | Class | Status |
|----------|-------|--------|
| Erlotinib | 1st-gen TKI | Approved |
| Gefitinib | 1st-gen TKI | Approved |
| Afatinib | 2nd-gen TKI | Approved |
| Dacomitinib | 2nd-gen TKI | Approved |
| Osimertinib | 3rd-gen TKI | Approved |

### Scaffold Frequency

| Scaffold | ChEMBL Count | Use |
|----------|--------------|-----|
| Quinazoline | 2,456 | Kinase inhibitors |
| Pyrimidine | 3,789 | Various targets |
| Pyrrolopyrimidine | 456 | Selective kinases |

## Scaffold Hopping Opportunities

### Ring Replacements

| Original | Bioisostere | Rationale |
|----------|-------------|-----------|
| Benzene | Pyridine | Add H-bond acceptor |
| Benzene | Thiophene | Slightly larger, polarizable |
| Pyridine | Pyrimidine | Add H-bond acceptor |
| Phenyl | Cyclohexyl | Remove aromaticity |

### Novel Scaffolds

**Identified 3 novel scaffolds** with similar topology:

1. **Indazole core**: 3 compounds
2. **Pyrrolopyrimidine**: 5 compounds
3. **Imidazopyridazine**: 2 compounds

## Property Comparison

| Property | Query | Mean (Similar) | Range |
|----------|-------|----------------|-------|
| MW | 393 | 420 | 350-480 |
| LogP | 3.2 | 3.5 | 2.8-4.2 |
| HBD | 1 | 1.2 | 1-2 |
| HBA | 3 | 3.5 | 2-5 |
| PSA | 72 | 78 | 65-95 |

## Recommendations

1. **Explore indazole compounds**: Novel scaffold, good properties
2. **Monitor pyrrolopyrimidines**: Emerging scaffold
3. **Consider scaffold hopping**: If IP crowded

Similarity Thresholds

Application Tanimoto Interpretation
Identical 1.0 Same compound
Very similar 0.9-1.0 Same analog series
Similar 0.7-0.9 Same scaffold
Related 0.5-0.7 Similar structure
Distant 0.3-0.5 Some similarity
Unrelated <0.3 Different chemotypes

Running Scripts

# Similarity search
python scripts/structure_search.py --query "SMILES" --threshold 0.7

# Scaffold analysis
python scripts/scaffold_analysis.py --input compounds.sdf --type murcko

# Scaffold hopping
python scripts/scaffold_hop.py --input compound.sdf --output hops.sdf

# Chemical space mapping
python scripts/chemical_space.py --library compounds.sdf --pca

Requirements

pip install rdkit pandas numpy scikit-learn

# Optional for visualization
pip install plotly seaborn matplotlib

Reference

Best Practices

  1. Use appropriate thresholds: 0.7 for similar compounds
  2. Consider scaffold: Different scaffold may have similar activity
  3. Check properties: Similar doesn't mean drug-like
  4. Validate experimentally: In-silico similarity needs confirmation
  5. Use multiple methods: Fingerprints + alignment for full picture

Common Pitfalls

Pitfall Solution
High similarity ≠ same activity Check bioactivity
Ignoring stereochemistry Use isomeric SMILES
Fingerprint bias Try multiple fingerprint types
Scaffold blindness Explicit scaffold analysis
Over-clustering Appropriate threshold selection