Feature trees: a new molecular similarity measure based on tree matching

Area of Science:

• Computational chemistry
• Cheminformatics
• Drug discovery

Background:

• Traditional molecular similarity methods often rely on linear representations like fingerprints.
• A more sophisticated molecular description is needed to capture complex structural relationships.

Purpose of the Study:

• To introduce a new method for evaluating molecular similarity using feature trees.
• To demonstrate the effectiveness of feature trees in identifying similar small organic compounds and predicting their binding modes.

Main Methods:

• A feature tree representation is generated for each molecule, detailing hydrophobic fragments, functional groups, and their linkages.
• Nodes in the feature tree are labeled with chemical properties.
• Subtree matching algorithms are developed and applied to compare feature trees.

Main Results:

• The feature tree approach successfully identified molecules belonging to the same class of inhibitors within a dataset of approximately 1000 compounds.
• For molecules with known binding modes, the algorithms achieved 61% compatibility with their relative orientation in the active site.
• Average computation time for pairwise comparison is approximately 50 ms.

Conclusions:

• Feature trees offer a robust method for molecular similarity assessment beyond linear fingerprints.
• The developed algorithms are efficient and accurate for identifying structurally related molecules and predicting binding poses.
• This method has potential applications in drug discovery and computational chemistry.