A novel graph-based similarity measure for 2D chemical structures

Area of Science:

• Computational Chemistry
• Cheminformatics
• Bioinformatics

Background:

• Measuring chemical compound similarity is crucial for drug discovery and understanding chemical reactions.
• Existing methods may not fully capture structural nuances relevant to biological function.
• Graph-based representations offer a powerful way to encode molecular structures.

Purpose of the Study:

• To develop and validate a novel graph-based method for quantifying 2D chemical compound similarity.
• To apply this similarity measure in clustering a large chemical database.
• To explore the biological and genomic implications of structurally similar compound clusters.

Main Methods:

• A graph-based similarity measure focusing on common subgraph edges, nodes, and connectivity was developed.
• This method was applied to over eleven thousand compounds from the KEGG/LIGAND database.
• A clustering algorithm was used in conjunction with the proposed similarity measure.

Main Results:

• The method successfully clustered compounds with high structural similarity.
• These clusters exhibited shared common names, participation in identical biological pathways, and similar enzyme requirements.
• A novel finding revealed sameness between pathway modules identified by structural clusters and those identified by genomic contexts (operon structures).

Conclusions:

• The proposed graph-based similarity measure is effective for grouping structurally related chemical compounds.
• Structural similarity correlates strongly with shared biological functions and pathway involvement.
• The study highlights an unexpected link between molecular structure similarity and gene organization in biological systems.