Enhancing fluorescence correlation spectroscopy with machine learning to infer anomalous molecular motion.

Area of Science:

• Biophysics
• Cellular Dynamics
• Data Analysis

Background:

• Molecular motion in cells often deviates from standard Brownian motion, termed anomalous diffusion.
• Fluorescence Correlation Spectroscopy (FCS) and Single-Particle Tracking (SPT) are key methods for studying cellular diffusion.
• Classical FCS analysis is limited to models with known analytical auto-correlation functions and requires long acquisition times.

Purpose of the Study:

• To develop a novel analysis approach for FCS that overcomes limitations of classical methods.
• To enable FCS to analyze complex anomalous diffusion models and reduce acquisition times.
• To enhance FCS's capability to study rapid changes in molecular motion within living cells.

Main Methods:

• Feature extraction from individual FCS recordings using an auto-correlation function estimator.
• Machine learning application to infer motion models and estimate motion parameters.
• Validation using simulated data and experimental recordings of fluorescent beads in glycerol solutions.

Main Results:

• The new FCS analysis approach successfully distinguishes various standard and anomalous diffusion models, including continuous-time random walks and fractal random walks.
• Performance is comparable to state-of-the-art SPT algorithms.
• The method accurately analyzes short FCS recordings and monitors rapid changes in motion parameters.
• Experimental results with fluorescent beads align with theoretical predictions from the Stokes-Einstein law.

Conclusions:

• The proposed machine learning-based FCS analysis significantly expands the capabilities of FCS.
• It provides analysis power similar to advanced SPT methods, enabling detailed study of anomalous diffusion.
• This approach allows for faster, more comprehensive investigation of molecular dynamics in biological systems.