Neural network informed photon filtering reduces fluorescence correlation spectroscopy artifacts.

Area of Science:

• Biophysics
• Microscopy Techniques
• Data Analysis

Background:

• Fluorescence correlation spectroscopy (FCS) is crucial for studying molecular dynamics.
• Artifacts like peak artifacts from bright, slow-moving clusters often render FCS measurements unanalyzable.
• Existing methods struggle with severe artifacts, leading to data loss.

Purpose of the Study:

• To develop an automated method for identifying and correcting peak artifacts in FCS time series.
• To restore the analyzability of FCS measurements compromised by artifacts.
• To provide a versatile tool for improving FCS data analysis.

Main Methods:

• Trained a one-dimensional U-Net model to automatically detect peak artifacts in fluorescence time series.
• Employed time-series editing to analyze purified, non-artifactual fluctuations.
• Validated the approach using simulations and two independent biological experiments.

Main Results:

• Successfully identified and corrected peak artifacts in simulated FCS data.
• Restored accurate transit time and particle number distributions in the presence of artifacts.
• Demonstrated the method's effectiveness in real biological experiments.

Conclusions:

• The developed automated pipeline effectively handles peak artifacts in FCS data.
• This approach makes previously unanalyzable data usable, enhancing molecular dynamics studies.
• The method is adaptable to other FCS artifacts, expanding its utility for researchers.