RCDPeaks: memory-efficient density peaks clustering of long molecular dynamics

Area of Science:

• Computational chemistry
• Biophysics
• Data analysis

Background:

• Density Peaks clustering is suitable for Molecular Dynamics (MD) simulations due to its handling of high-density and distant cluster centers.
• Existing Density Peaks implementations have quadratic memory complexity, limiting their use to short MD trajectories.

Purpose of the Study:

• To introduce DP+ and RCDPeaks, novel implementations of Density Peaks for efficient MD data analysis.
• To overcome the memory limitations of traditional Density Peaks algorithms for large-scale MD simulations.

Main Methods:

• Developed DP+, an exact implementation of Density Peaks with reduced RAM consumption.
• Created RCDPeaks, a refined variant of Density Peaks utilizing DP+.
• Demonstrated RCDPeaks' capability to cluster a million-frame trajectory using minimal RAM.

Main Results:

• RCDPeaks clustered a one-million frame MD trajectory using less than 4.5 GB of RAM.
• This represents a significant reduction compared to over 2 TB required by existing alternatives.
• RCDPeaks offers automatic parameter selection, center candidate screening, and cluster refinement.

Conclusions:

• DP+ and RCDPeaks provide a computationally efficient solution for clustering large MD datasets.
• RCDPeaks significantly lowers the memory and time requirements for analyzing MD simulation trajectories.
• The RCDPeaks software is publicly available on GitHub for broader scientific use.