An automated clustering workflow for molecular simulation data

Area of Science:

• Computational chemistry
• Biophysics
• Data science

Background:

• Modern molecular simulations generate massive datasets requiring efficient analysis.
• Dimensionality reduction and clustering are key techniques for understanding complex systems.
• Identifying metastable states is crucial for molecular dynamics studies.

Purpose of the Study:

• To provide a refined and automated clustering workflow for large molecular simulation data.
• To improve the efficiency and quality of cluster analysis in computational studies.
• To apply the workflow to atomistic simulations of the ubiquitin protein.

Main Methods:

• Combines dimensionality reduction (cc_analysis, EncoderMap) with clustering (HDBSCAN).
• Utilizes a root mean square deviation-based sorting criterion for structure homogeneity.
• Employs an iterative approach for efficient categorization of large datasets.

Main Results:

• The automated workflow successfully categorizes large datasets into structurally homogeneous clusters.
• The refined method demonstrates improvements in cluster quality and homogeneity.
• Application to ubiquitin simulations highlights the workflow's effectiveness.

Conclusions:

• The automated clustering workflow offers a powerful tool for analyzing large molecular simulation datasets.
• This method enhances the identification of metastable states and structural features.
• The refined approach provides significant improvements over previous methods.