Ligand-Based Virtual Screening Using Graph Edit Distance as Molecular Similarity Measure

Area of Science:

• Computational Chemistry
• Cheminformatics
• Drug Discovery

Background:

• Molecular similarity calculations are crucial for ligand-based virtual screening.
• Current methods often convert complex graph structures into simpler vector representations.
• This can lead to a loss of detailed structural and property information.

Purpose of the Study:

• To investigate the effectiveness of a graph-only molecular comparison using extended reduced graphs and graph edit distance.
• To assess this approach for ligand-based virtual screening applications.
• To compare its performance against existing methods.

Main Methods:

• Utilized extended reduced graphs with pharmacophore-type node descriptions.
• Employed graph edit distance for molecular similarity calculations.
• Applied screening and statistical tools from a virtual screening benchmarking platform and RDKit.
• Validated on six public datasets (DUD-E, MUV, GLL&GDD, CAPST, NRLiSt BDB, ULS-UDS).

Main Results:

• The graph edit distance method demonstrated stable and superior performance compared to previous reduced graph methods.
• The proposed graph-only approach outperformed array-based molecular similarity methods in most cases.
• The method successfully identified bioactivity similarities across structurally diverse molecules.

Conclusions:

• Extended reduced graphs combined with graph edit distance offer a robust method for molecular similarity.
• This approach is effective for ligand-based virtual screening.
• The technique has broad applications in identifying bioactivity in diverse chemical compound libraries.