Clustering one million molecular structures on GPU within seconds.

Area of Science:

• Computational Biology
• Structural Biology
• Biophysics

Background:

• Structure clustering is crucial for analyzing molecular dynamics simulations in life sciences.
• Existing tools often lack support for GPU acceleration and the root mean square deviation (RMSD) metric, leading to lengthy computation times.
• Efficient analysis of large conformational datasets is essential for understanding molecular behavior.

Purpose of the Study:

• To develop and present a novel, efficient software for structure clustering using graphics processing units (GPUs).
• To enable GPU-accelerated clustering analysis with the root mean square deviation (RMSD) metric.
• To provide a powerful tool for rapidly identifying representative molecular structures.

Main Methods:

• Developed custom code for multi-threaded, multi-GPU parallel processing.
• Implemented clustering analysis utilizing the root mean square deviation (RMSD) metric.
• Tested the program on a large dataset (1.4 million snapshots) from an enhanced sampling simulation of a protein fragment (Pin1 WW domain mutant).

Main Results:

• The program demonstrates high efficiency, completing the clustering of 1 million snapshots in seconds using two NVIDIA RTX4090 GPUs.
• Achieved speedups hundreds of times faster than traditional central processing unit (CPU) methods.
• Successfully handled large datasets distributed widely in conformational space.

Conclusions:

• The developed GPU-accelerated software offers a significant performance improvement for structure clustering.
• This tool can drastically reduce the time required to extract representative molecular states from extensive simulation data.
• It serves as a powerful asset for researchers in life sciences dealing with large-scale structural analyses.