The algebraic extended atom-type graph-based model for precise ligand-receptor binding affinity prediction.

Area of Science:

• Computational chemistry
• Structural biology
• Drug discovery

Background:

• Accurate prediction of ligand-receptor binding affinity is vital for structure-based drug design.
• Machine learning-based scoring functions have advanced predictions but struggle with complex molecular interactions.

Purpose of the Study:

• Introduce the AGL-EAT-Score, a novel scoring function for predicting ligand-receptor binding affinity.
• Improve the accuracy and reliability of drug design tools.

Main Methods:

• Integrate extended atom-type multiscale weighted colored subgraphs with algebraic graph theory.
• Utilize eigenvalues and eigenvectors of graph Laplacian and adjacency matrices.
• Perform comprehensive similarity analysis of protein sequence, ligand structure, and binding site.

Main Results:

• AGL-EAT-Score demonstrated notable accuracy on benchmark datasets (CASF-2016, CASF-2013, Cathepsin S).
• Outperformed existing traditional and machine learning-based scoring functions.
• Effectively captured intricate protein-ligand interactions using extended atom types.

Conclusions:

• AGL-EAT-Score offers a significant advancement in structure-based drug design.
• Provides a robust and systematic tool to refine and accelerate the drug discovery process.
• Addresses challenges of dataset bias and over-representation in predictive models.