MP-NeRF: A massively parallel method for accelerating protein structure reconstruction from internal coordinates.

Area of Science:

• Computational Biology
• Biophysics
• Structural Bioinformatics

Background:

• Protein coordinate conversion is crucial for molecular dynamics simulations and machine learning.
• Current methods using the Natural extension of Reference Frame (NeRF) algorithm can be computationally intensive.
• Efficiency bottlenecks in NeRF limit the scale and speed of biomolecular modeling pipelines.

Purpose of the Study:

• To develop a massively parallel implementation of the NeRF algorithm for protein coordinate conversion.
• To significantly enhance the speed and scalability of converting between internal and Cartesian protein coordinates.
• To provide a reusable and user-friendly open-source tool for the scientific community.

Main Methods:

• A novel, massively parallel NeRF algorithm was developed.
• The conversion process was divided into three phases: monomer backbone composition, backbone subunit assembly, and sidechain elongation.
• Computations were batched into efficient matrix operations for parallel processing.

Main Results:

• Achieved speedups ranging from 400× to 1200× compared to the state-of-the-art.
• Demonstrated significant performance gains dependent on polymer length.
• Developed an open-source Python package for broad accessibility.

Conclusions:

• The massively parallel NeRF implementation dramatically accelerates protein coordinate conversions.
• This advancement is expected to benefit molecular dynamics simulations and machine learning models.
• The open-source nature promotes wider adoption and further development in computational structural biology.