DeepRank

DeepRank analyzes protein-protein interaction (PPI) interfaces using three-dimensional convolutional neural networks (CNNs) to extract spatial features for classification and regression tasks such as distinguishing biological versus crystallographic PPIs and ranking docking models.

Key Features:

• 3D Convolutional Neural Networks: DeepRank employs three-dimensional convolutional neural networks (CNNs) to capture spatial features of protein complexes by mapping interfaces onto 3D grids.
• Configurable Architecture: The framework provides a configurable model architecture to tailor input features and network parameters for diverse structural biology tasks.
• Scalability for Large Datasets: DeepRank supports efficient training on datasets comprising millions of PPIs to enable large-scale data mining.
• Classification and Regression Support: The framework supports both classification and regression tasks, enabling distinction of biological versus crystallographic PPIs and scoring or ranking of docking models.
• APIs for Data Handling: DeepRank exposes APIs for pre-processing, feature computation, and model evaluation.

Scientific Applications:

• Classification of Biological vs. Crystallographic PPIs: DeepRank distinguishes biological interfaces from crystallographic contacts using learned spatial features.
• Ranking of Docking Models: DeepRank ranks docking models by predicted relevance or quality based on interface features.

Methodology:

DeepRank maps PPI data onto 3D grids that serve as inputs to train three-dimensional CNNs, and its APIs support pre-processing, feature computation, and model evaluation.