High-fidelity predictions of diffusion in the brain microenvironment.

Area of Science:

• Neuroscience
• Biophysics
• Computational Biology

Background:

• Multiple-particle tracking (MPT) characterizes biological microenvironments like the brain's extracellular space (ECS).
• Machine learning (ML) has been used to analyze MPT data for predicting diffusion modes and biological variables.
• Investigating brain ECS changes due to injury is crucial for understanding neurological conditions.

Purpose of the Study:

• Develop and validate an ML pipeline to predict and investigate brain ECS alterations caused by injury.
• Assess the pipeline's effectiveness across different experimental conditions, including age, region, and enzymatic degradation.
• Identify novel, biologically relevant features of nanoparticle diffusion indicative of injury.

Main Methods:

• Utilized a supervised classification ML pipeline with feature importance calculations.
• Validated the pipeline on MPT data sets related to age, region, and enzymatic ECS structure degradation.
• Employed linear mixed effects models for initial comparisons between healthy and injured brain tissue (oxygen-glucose deprivation - OGD).
• Applied ML to predict injury states (control, 0.5h OGD, 1.5h OGD) using MPT features.

Main Results:

• Achieved high accuracy in predicting age (86%), brain regions (90%), and enzyme-treated tissue (69%).
• Successfully predicted injury states in cortex (59%) and striatum (66%) using MPT features.
• Identified nonlinear relationships between trajectory features, surpassing traditional linear models in revealing injury-specific changes.

Conclusions:

• The developed ML pipeline effectively analyzes MPT data across diverse experimental conditions.
• ML applied to MPT data can uncover unique, biologically relevant features of nanoparticle diffusion related to brain injury.
• This approach offers a powerful tool for investigating complex biological microenvironments and injury mechanisms.