GradPose

GradPose performs gradient-descent-based superimposition of protein structures to enable fast, large-scale comparison of conformations from simulations such as molecular dynamics and docking.

Key Features:

• Gradient Descent Optimization: Optimizes rotation quaternions via gradient descent to align protein structures.
• Handling Structural Variability: Accommodates insertions and deletions relative to a reference structure.
• Scalability and Performance: Capable of superimposing thousands to millions of structures on standard hardware and leverages multiple CPU cores with optional CUDA acceleration.
• Efficiency Gains: Reports speed improvements of approximately 2–65× and memory reductions of approximately 1.7–48× compared to traditional methods, with larger gains for bigger proteins.
• Limitations: Requires a predetermined residue-residue correspondence and may be outperformed by traditional methods for very small proteins (~20 residues).

Scientific Applications:

• Ensemble analysis of molecular dynamics and docking simulations: Enables large-scale superimposition for comparative analysis of conformational ensembles generated by molecular dynamics and docking.
• Comparison and clustering of protein structures: Facilitates structure-to-structure comparisons and downstream clustering of large structural datasets.
• Study of protein dynamics and interactions at scale: Supports exploration of conformational variability and interaction conformations across extensive simulation datasets.

Methodology:

Optimizes rotation quaternions using gradient descent and can utilize multiple CPU cores and CUDA acceleration.