Measurement of anomalous diffusion using recurrent neural networks

Area of Science:

• Physics
• Biophysics
• Data Science

Background:

• Anomalous diffusion describes physical and biological processes where mean squared displacement (MSD) growth deviates from linear time dependence.
• Traditional MSD-based exponent estimation struggles with limited experimental data points.

Purpose of the Study:

• To demonstrate recurrent neural networks (RNNs) for efficient anomalous diffusion characterization.
• To showcase RNNs' ability to determine anomalous diffusion exponents from short, potentially irregularly sampled trajectories.
• To apply RNNs to complex intermittent diffusion systems.

Main Methods:

• Utilized recurrent neural networks (RNNs) to analyze trajectory data.
• Compared RNN performance against standard MSD-based methods for exponent estimation.
• Applied RNNs to experimental data from subdiffusive colloids and superdiffusive microswimmers.

Main Results:

• RNNs accurately determine anomalous diffusion exponents from single, short trajectories, surpassing MSD methods with limited data.
• RNNs successfully analyzed trajectories sampled at irregular time intervals.
• RNNs estimated switching times and exponents in intermittent anomalous diffusion systems.

Conclusions:

• RNNs offer a powerful and versatile tool for analyzing anomalous diffusion across various physical and biological systems.
• The proposed RNN approach enhances the analysis of complex diffusion dynamics, especially under experimental constraints.