Leveraging Machine Learning-Guided Molecular Simulations Coupled with Experimental Data to Decipher Membrane Binding...

Area of Science:

• Computational chemistry and biophysics
• Molecular dynamics simulations
• Neuroscience

Background:

• Understanding molecular interactions with cell membranes is crucial for drug delivery and understanding neurodegenerative diseases.
• Estimating the free energy landscape of solute binding to biological membranes remains a significant computational challenge.

Purpose of the Study:

• To determine the free energy landscape of membrane insertion for aminosterol compounds trodusquemine (TRO) and squalamine (SQ).
• To develop and validate a computational strategy for characterizing solute-membrane interactions and predicting binding affinities.

Main Methods:

• Leveraging the Deep-TICA approach, integrating Time-lagged Independent Component Analysis (TICA) with neural networks (NN).
• Utilizing on-the-fly probability enhanced sampling (OPES) for free energy landscape estimation.
• Simulating aminosterol insertion into a neuron-like lipid bilayer model.

Main Results:

• The Deep-TICA strategy effectively generated a collective variable for describing solute absorption into membranes.
• Accurate prediction of membrane binding affinities for TRO and SQ, aligning closely with experimental data from large unilamellar vesicles (LUVs).
• Exhaustive characterization of the aminosterol entry pathway into the lipid bilayer was achieved.

Conclusions:

• The computational protocol provides a robust method for investigating membrane binding processes and predicting binding affinities.
• Findings enhance understanding of aminosterol entry mechanisms and their protective role against neurodegeneration.
• The Deep-TICA approach shows significant potential for future studies on solute-membrane interactions.