Comparing Molecular Patterns Using the Example of SMARTS: Theory and Algorithms

Area of Science:

• Computational chemistry
• Cheminformatics

Background:

• Molecular patterns are crucial for filtering compounds in drug design, identifying structural properties linked to undesirable traits like toxicity.
• Existing methods struggle to analytically compare large sets of complex structural filters.

Purpose of the Study:

• To develop a novel, generic approach for comparing molecular patterns and analyzing structural filter collections.
• To introduce a tool for automated pattern hierarchy derivation and similarity searching.

Main Methods:

• Chemically inspired fingerprints were developed for pattern nodes and edges to create a comparable pattern representation.
• A maximum common subgraph algorithm was applied to annotated pattern graphs to calculate pattern inclusion and similarity.
• The approach was implemented in a tool called SMARTScompare for the SMARTS language.

Main Results:

• The developed algorithm enables the calculation of pattern inclusion and similarity.
• SMARTScompare can automatically derive pattern hierarchies and search large pattern collections for related patterns.
• The tool's capabilities were demonstrated on a substantial dataset of real-world SMARTS patterns.

Conclusions:

• This work presents the first algorithm capable of chemical pattern analytics for deriving hierarchies and performing similarity searches.
• SMARTScompare offers a powerful new capability for analyzing and managing collections of structural filters in molecular design.
• The approach facilitates more efficient and systematic compound filtering and molecular design processes.