Learning Continuous 2D Diffusion Maps from Particle Trajectories without Data Binning.

Area of Science:

• Biophysics
• Computational Biology
• Membrane Dynamics

Background:

• Diffusion coefficients vary spatially within cellular membranes, offering insights into local properties like composition and stiffness.
• Quantifying these variations requires advanced analytical methods beyond traditional approaches.
• Single-molecule imaging provides rich data but demands sophisticated analysis for diffusion mapping.

Approach:

• A Bayesian framework is employed, utilizing Gaussian Process (GP) priors to model diffusion coefficient spatial maps.
• Inducing point methods are leveraged for computational efficiency, particularly with non-conjugate likelihood-prior pairs.
• The method is validated on both synthetic datasets and real-world live-cell single-molecule imaging data.

Key Points:

• The developed tool generates continuous spatial maps of diffusion coefficients across membranes.
• Uncertainties in the diffusion coefficient maps are rigorously quantified.
• The method operates in an unsupervised manner, eliminating the need for data binning.

Conclusions:

• This unsupervised Bayesian approach accurately maps diffusion coefficients and their uncertainties across cellular membranes.
• The technique offers a powerful new tool for investigating local membrane heterogeneity and dynamics.
• It advances the understanding of membrane biophysics by providing high-resolution diffusion information.