MP-NeRF

MP-NeRF converts internal coordinates to Cartesian coordinates for polymers using a massively parallel Natural Extension Reference Frame (NeRF) implementation to accelerate molecular dynamics simulations and machine-learning-based protein structure prediction.

Key Features:

• Massively Parallel Implementation: Introduces a massively parallel version of the Natural Extension Reference Frame (NeRF) algorithm to accelerate coordinate conversion.
• Speed Improvements: Achieves speedups of approximately 400–1,200× compared to previous state-of-the-art methods depending on polymer length.
• Efficient Conversion Phases: Performs parallel composition of monomer backbone, assembly of backbone subunits, and parallel elongation of sidechains as distinct phases of conversion.
• Optimized Matrix Operations: Batches computations into a minimal number of efficient matrix operations to maximize computational efficiency and minimize overhead.

Scientific Applications:

• Molecular dynamics simulations: Accelerates generation of Cartesian coordinates from internal representations for use in molecular dynamics workflows.
• Machine learning-based protein structure prediction: Provides fast conversion between coordinate representations to support machine-learning models that predict protein structures.
• Protein folding, function, and interaction studies: Enables more rapid iterations and larger-scale analyses to facilitate studies of protein folding, function, and interactions.

Methodology:

Divides the conversion into parallel composition of monomer backbone, assembly of backbone subunits, and parallel elongation of sidechains executed concurrently, and batches computations into a minimal number of efficient matrix operations.