An electron-density point-cloud framework for robust protein-ligand interaction prediction.

Area of Science:

• Computational chemistry
• Structural biology
• Drug discovery

Background:

• Accurate protein-ligand affinity prediction is crucial for drug discovery.
• Current methods often rely on high-resolution 3D structures, limiting their applicability to low-resolution or predicted protein models.
• Robustness is needed to handle variations in structural data quality.

Purpose of the Study:

• To develop a novel framework, E-CloudBind, for robust protein-ligand affinity prediction.
• To overcome the limitations of structure-based methods that require sub-ångström accuracy.
• To enable accurate predictions using heterogeneous structural data, including low-resolution and predicted structures.

Main Methods:

• E-CloudBind fuses electron-density point clouds of ligands with intrinsic molecular graphs of proteins.
• Ligand electron densities are generated using semi-empirical quantum calculations.
• Protein pockets are represented by van der Waals-guided Gaussian point clouds, tolerating coordinate noise.
• A heterogeneous graph neural network integrates point-cloud features and graph features for affinity regression.

Main Results:

• E-CloudBind achieves performance matching or exceeding leading baselines across various benchmarks, including PDBbind.
• The framework demonstrates markedly reduced sensitivity to protein structure resolution and experimental versus predicted structures.
• Case studies show atom-level interpretability and successful application in large-scale virtual screening.

Conclusions:

• E-CloudBind provides a robust and accurate method for protein-ligand affinity prediction, independent of precise 3D coordinates.
• The framework's ability to handle heterogeneous structural data opens new possibilities for structure-based drug design.
• By decoupling interaction learning from exact coordinates, E-CloudBind enhances the applicability of computational methods in real-world scenarios.