AGL-Score: Algebraic Graph Learning Score for Protein-Ligand Binding Scoring, Ranking, Docking, and Screening

Area of Science:

• Computational chemistry
• Graph theory
• Machine learning

Background:

• Traditional algebraic graph theory models struggle with predicting biomolecular properties.
• The spectrum-geometry relationship in graphs remains a long-standing challenge.

Purpose of the Study:

• To introduce novel algebraic graph learning score (AGL-Score) models for enhanced biomolecular property prediction.
• To bridge the gap between spectral graph theory and geometric inference in biomolecular studies.

Main Methods:

• Developed AGL-Score models utilizing multiscale weighted colored subgraphs and graph invariants (Laplacian, pseudo-inverse, adjacency matrices).
• Integrated AGL-Score models with machine learning algorithms for predicting macroscopic biomolecular properties from low-dimensional graph representations.
• Validated models using CASF-2007, CASF-2013, and CASF-2016 benchmark datasets.

Main Results:

• AGL-Score models demonstrated superior performance in scoring, ranking, docking, and screening power compared to state-of-the-art methods.
• The models effectively encoded high-dimensional physical and biological information into low-dimensional representations.
• Validated the efficacy of spectral graph theory in inferring geometric properties.

Conclusions:

• AGL-Score models represent a significant advancement in applying graph theory and machine learning to biomolecular studies.
• Machine learning and spectral graph theory are powerful tools for molecular docking and virtual screening.
• The study successfully inferred geometric properties from spectral data in biomolecular contexts.