A survey of algorithms for transforming molecular dynamics data into metadata for in situ analytics based on machine...

Area of Science:

• Computational chemistry and materials science
• Biophysics and structural biology
• High-performance computing and data analytics

Background:

• Molecular dynamics (MD) simulations are crucial in chemistry, materials science, and biology, generating vast datasets.
• Current analysis methods struggle with the scale of data from next-generation supercomputers.
• Traditional data representations are inadequate for real-time analysis of extensive MD trajectories.

Purpose of the Study:

• To present and survey three algorithms for transforming atomic-level MD simulation data into metadata representations.
• To enable in situ analytics using machine learning on large-scale MD simulation data.
• To facilitate efficient data preparation for advanced analysis in scientific computing.

Main Methods:

• Development and application of three distinct algorithms for data transformation.
• Conversion of molecular snapshots into concise, meaningful metadata.
• Integration of metadata with machine learning methods for runtime analysis.

Main Results:

• Demonstrated the efficacy of the algorithms in creating suitable metadata representations.
• Enabled local, on-the-fly data transformation for large MD trajectories.
• Facilitated in situ machine learning analysis for complex biological systems.

Conclusions:

• The proposed algorithms are essential for preparing data for in situ analysis of large molecular dynamics simulations.
• These methods support the efficient use of high-performance computing resources for scientific discovery.
• The approach is applicable to diverse fields including drug design, protein folding, and protein engineering.