Viscosity Measurement in Biocondensates Using Deep-Learning-Assisted Single-Particle Rotational Analysis

Area of Science:

• Biophysics
• Biomolecular Condensates
• Nanotechnology

Background:

• Viscoelastic characterization is crucial for studying biomolecular condensates.
• Single-nanorod rotational diffusion analysis offers quantitative viscosity measurements.
• High noise and background in imaging complicate accurate diffusivity extraction.

Purpose of the Study:

• To develop a novel deep learning (DL) method for analyzing single nanorod rotational tracking.
• To enhance the accuracy and range of viscosity measurements in challenging imaging conditions.

Main Methods:

• Developed a frequency-domain deep learning approach for nanorod rotational tracking.
• Utilized synthesized Brownian rotational time-series data for recurrent neural network (RNN) training.
• Implemented a data preprocessing module to mitigate noise effects.

Main Results:

• The proposed DL method demonstrated higher accuracy than traditional curve-fitting techniques.
• Achieved a wider detection range for viscosity measurements.
• Successfully validated using experimental data from gold nanorods in glycerol and PGL droplets.

Conclusions:

• The frequency-domain DL method provides a robust and accurate approach for viscosity measurement.
• This technique effectively addresses challenges posed by high background and noise in imaging.
• Potential applications include *in vitro* and *in vivo* viscosity measurements of diverse biomolecular condensates.