A multilayer dynamic perturbation analysis method for predicting ligand-protein interactions.

Area of Science:

• Computational Biology
• Structural Biology
• Bioengineering

Background:

• Ligand-protein interactions are fundamental to protein function.
• AI-driven structure prediction necessitates accurate, structure-based natural ligand prediction methods.
• Existing methods struggle with predicting ligand-binding regions and spatial ligand extension.

Purpose of the Study:

• To develop an accurate and rapid structure-based method for predicting natural ligand-binding regions.
• To address the limitations of current methods in identifying ligand-binding areas and ligand spatial distribution within pockets.
• To leverage protein structure for predicting ligand interactions.

Main Methods:

• Proposed a multilayer dynamics perturbation analysis (MDPA) method, an extension of fast dynamic perturbation analysis (FDPA).
• MDPA identifies ligand-binding regions by analyzing areas causing significant changes in protein conformational dynamics.
• The method relies solely on protein structure data.

Main Results:

• MDPA achieved an average ligand-binding site prediction Matthews coefficient of 0.40 on a standard validation dataset.
• Achieved prediction precision of at least 50% for 71% of cases.
• Demonstrated significant overlap (>=50%) between predicted regions and natural ligands in 80% of cases, outperforming other state-of-the-art methods.

Conclusions:

• MDPA is an effective structure-based tool for detecting ligand-binding regions on protein surfaces.
• Identified subtle protein pocket interactions significantly impacting protein dynamics.
• Provides a foundational tool for natural ligand detection and protein functional annotation, with source code available.