Recognition of Coexisting Phases in Model Membranes via an Unsupervised Method

Area of Science:

• Biophysics
• Computational Biology
• Membrane Biophysics

Background:

• Cell membranes exhibit lateral heterogeneity due to lipid phase separation.
• Molecular dynamics (MD) simulations provide atomistic details of lipid phases.
• Identifying lipid phases from MD simulations is computationally challenging.

Purpose of the Study:

• To develop an unsupervised method for accurate lipid phase recognition in phase-separated lipid bilayers.
• To provide a robust framework for analyzing lipid heterogeneity in complex membrane systems.

Main Methods:

• Discretizing the membrane plane into pixels and calculating local lipid packing degree (atomic density).
• Utilizing a Gaussian mixture model (GMM) to determine a threshold for phase assignment.
• Mapping identified phases back to individual lipids within the bilayer.

Main Results:

• The GMM-based method accurately identifies coexisting lipid phases across diverse bilayer compositions and temperatures.
• This approach outperforms the hidden Markov model (HMM) in accuracy and robustness.
• Phase recognition relies solely on intrinsic bilayer properties (lipid packing degree), requiring no temporal or labeled data.

Conclusions:

• The developed method offers a robust and accurate framework for identifying and tracking lipid phases in simulations.
• This technique is broadly applicable to various systems, including protein-containing bilayers and pre-equilibration dynamics.
• It enhances the analysis of cell membrane lateral heterogeneity and phase transitions.