Ligand Unbinding Pathway and Mechanism Analysis Assisted by Machine Learning and Graph Methods.

Area of Science:

• Computational Chemistry
• Molecular Dynamics
• Biophysics

Background:

• Understanding protein-ligand unbinding is crucial for drug discovery.
• Biased simulations are often used to explore unbinding pathways.
• Analyzing complex simulation data to extract meaningful pathways remains challenging.

Purpose of the Study:

• To present two novel computational methods for analyzing protein-ligand unbinding mechanisms.
• To cluster simulation trajectories into representative unbinding paths.
• To identify challenges in detecting reaction coordinates for unbinding path analysis.

Main Methods:

• Contact principal component analysis (CPCA) for dimensionality reduction and machine learning-based trajectory clustering.
• Neighbor-net algorithm for trajectory clustering based on pairwise Euclidean distances, accounting for data bias.
• Analysis of a complex case involving an intraligand hydrogen bond as a reaction coordinate.

Main Results:

• The two presented methods effectively cluster simulation trajectories to represent unbinding paths.
• The neighbor-net algorithm demonstrates superiority over traditional dendrogram construction for biased trajectory data.
• Identifying the correct reaction coordinate, such as a specific hydrogen bond, presents significant challenges.

Conclusions:

• The developed methods provide powerful tools for dissecting protein-ligand unbinding mechanisms from simulation data.
• Accurate identification of unbinding pathways and reaction coordinates is essential for understanding molecular interactions.
• Further development is needed to address complexities in reaction coordinate detection for intricate systems.