Reasoning about molecular similarity and properties

Area of Science:

• Computational Chemistry
• Cheminformatics
• Molecular Modeling

Background:

• Molecular similarity is crucial for drug discovery and understanding biological properties.
• Current methods often rely on 2D representations, limiting the use of complex 3D molecular data.
• 3D surface-based molecular representations offer high descriptive power but are computationally challenging.

Purpose of the Study:

• To develop a robust method for determining molecular similarity using 3D surface-based representations.
• To address the limitations of existing 2D graph-based similarity approaches.
• To facilitate efficient similarity calculations that incorporate molecular geometry and fields.

Main Methods:

• Proposed an intrinsic, spherical representation mapping molecular surface points to a standard sphere.
• Captured molecular geometry and fields as distributions on the sphere's surface.
• Utilized a novel histogram-intersection formulation to compute similarity between property distributions.

Main Results:

• The spherical representation effectively captures molecular properties and facilitates similarity computation.
• The method is robust to noise, eliminates the need for pose optimization, and handles conformational variations.
• Demonstrated high retrieval performance and effectiveness in structure-activity modeling.

Conclusions:

• The proposed spherical representation and histogram-intersection method offer a powerful and efficient approach to molecular similarity assessment.
• This technique enhances the utility of 3D surface-based molecular representations in computational chemistry and drug discovery.
• The method shows significant advantages over existing techniques in terms of robustness, efficiency, and applicability.