PointDE: Protein Docking Evaluation Using 3D Point Cloud Neural Network.

Area of Science:

• Computational Biology
• Structural Biology
• Bioinformatics

Background:

• Protein-protein interactions (PPIs) are crucial for cellular functions.
• Determining protein complex structures aids in understanding PPI mechanisms.
• Protein-protein docking models complex structures but struggles with decoy selection.

Purpose of the Study:

• To develop an advanced method for evaluating protein-protein docking models.
• To improve the selection of accurate, near-native decoys from docking simulations.
• To enhance the understanding of protein-protein interaction mechanisms.

Main Methods:

• Developed PointDE, a 3D point cloud neural network for docking evaluation.
• Transformed protein structures into point cloud representations.
• Utilized a novel grouping mechanism within a state-of-the-art point cloud network architecture.

Main Results:

• PointDE demonstrated superior performance compared to existing deep learning methods on public datasets.
• The method showed strong performance on a new dataset of antibody-antigen complexes.
• PointDE effectively captures geometric information and learns from protein interfaces.

Conclusions:

• PointDE offers a robust and effective solution for evaluating protein-protein docking models.
• The method shows promise for diverse protein structure types, including antibody-antigen complexes.
• PointDE will facilitate deeper insights into the mechanisms of protein-protein interactions.