InDeep

InDeep predicts functional binding sites on protein structures using 3D fully convolutional neural networks to support drug design targeting protein–protein interactions (PPIs).

Key Features:

• 3D fully convolutional neural networks: Uses 3D fully convolutional neural networks for spatially resolved prediction of functional binding sites.
• Functional binding site prediction: Predicts functional binding sites that can host protein epitopes or future drugs.
• Training data: Trained on a curated dataset of protein–protein interactions (PPIs).
• Applicability to structures and dynamics: Applies predictions to experimental protein structures and along molecular dynamics trajectories.
• Benchmark performance: Demonstrates superior performance versus state-of-the-art predictors for ligandable binding sites on PPI and conventional drug targets in benchmarking.
• PPI interface focus: Targets binding pockets at or near PPI interfaces relevant to drug design.

Scientific Applications:

• Drug design for PPIs: Supports identification and prioritization of binding pockets for designing modulators of protein–protein interactions, including infectious disease targets.
• Epitope identification: Identifies protein epitopes by predicting functional binding sites.
• Dynamic pocket analysis: Enables analysis of ligandability along molecular dynamics trajectories to capture transient binding pockets.
• Target and hit prioritization: Ranks ligandable binding sites for both PPI targets and conventional drug targets based on model predictions and benchmarking.

Methodology:

Trains 3D fully convolutional neural networks on a curated PPI dataset and applies the models to experimental protein structures and molecular dynamics trajectories; performance was evaluated by benchmarking against state-of-the-art ligandable binding site predictors.