iScore

iScore ranks protein–protein docking models to identify near‑native conformations using random‑walk graph kernels and support vector machine classification.

Key Features:

• Graph kernel–based scoring function: Evaluates docking models by comparing interface graphs using random‑walk graph kernels against a training set.
• Support Vector Machine (SVM) classification: Uses a support vector machine to classify and rank protein–protein interfaces based on graph‑kernel similarities.
• MPI-based automation and scalability: Provides executable scripts that automate workflows and distribute computations across processes via Message Passing Interface (MPI).
• GPU acceleration (CUDA): Offloads computationally intensive kernel computations to GPUs through CUDA kernels to accelerate processing.
• Training and prediction binaries: Includes the binaries iscore.train and iscore.predict for model training and prediction of near‑native conformations.

Scientific Applications:

• Docking model ranking: Prioritizes near‑native protein–protein docking conformations among large sets of generated models.
• Structural modeling of complexes: Supports computational docking studies for modeling three‑dimensional structures of protein–protein complexes.
• Functional analysis of interactions: Aids structural biologists and bioinformaticians in analyzing molecular interactions and inferring protein functions within cellular contexts.

Methodology:

Construct interface graphs from protein–protein complexes, compute random‑walk graph kernels to assess interface similarities, train and apply a support vector machine for classification, and use MPI for task distribution with optional CUDA kernels for GPU acceleration.