ML-DBSCAN

ML-DBSCAN performs hierarchical density-based clustering of high-dimensional molecular dynamics (MD) simulation data to identify metastable states and map free energy landscapes.

Key Features:

• Hierarchical Clustering: Integrates clustering results across multiple density levels to capture both coarse-grained and fine-grained structures within MD data.
• Free Energy Landscape Analysis: Maps free energy landscapes and identifies metastable states, detecting high-population regions at low resolution and revealing structural details at higher resolution.
• Resolution Flexibility: Operates across varying density/resolution levels to allow analysis from broad patterns to intricate structural details.
• GPU Implementation: Provides a GPU implementation that accelerates processing, achieving up to two orders of magnitude speedup over the CPU implementation for larger MD datasets.
• High-dimensional MD Data Support: Analyzes high-dimensional datasets generated from molecular dynamics simulations.

Scientific Applications:

• Metastable State Identification: Aids identification and characterization of metastable states relevant to protein folding, conformational changes, and ligand binding.
• Structural Detailing: Provides insights into the structural details of free energy landscapes to support understanding of dynamic mechanisms.
• Versatility Across Systems: Demonstrated on systems including a 2D-potential, alanine dipeptide, β-hairpin Tryptophan Zipper 2 (Trpzip2), Human Islet Amyloid Polypeptide (hIAPP), and Maltose Binding Protein (MBP).

Methodology:

ML-DBSCAN applies Multi-Level Density-Based Spatial Clustering of Applications with Noise (ML-DBSCAN) by performing density-based clustering across multiple density levels to construct a hierarchical representation of the free energy landscape from MD data without requiring prior knowledge of structural details or dynamic mechanisms.