ExoFILT: Transfer learning for robust and accelerated analysis of exocytosis single-particle tracking data

Area of Science:

• Cell Biology
• Biophysics
• Computational Biology

Background:

• Quantitative characterization of protein dynamics in constitutive exocytosis is crucial for molecular understanding.
• Single-particle tracking (SPT) is a key technique for measuring protein dynamics in live cells.
• Manual annotation of SPT data for exocytic events is time-consuming, biased, and hinders reproducibility.

Purpose of the Study:

• To develop and validate a deep learning-based classifier, ExoFILT, for automated identification of exocytic events in SPT data.
• To improve the throughput and consistency of exocytic event analysis.
• To enable systematic quantification of temporal relationships between proteins involved in exocytosis.

Main Methods:

• Development of ExoFILT, a deep learning classifier utilizing transfer learning.
• Training ExoFILT on both simulated and experimental SPT data, referencing the exocyst complex.
• Application of ExoFILT to dual-color time-lapse movies for analyzing protein dynamics during exocytosis.

Main Results:

• ExoFILT reduces manual annotation time by tenfold compared to traditional methods.
• The classifier enhances measurement consistency across different researchers.
• Systematic quantification revealed distinct subpopulations of exocytic events with varying molecular compositions, such as the presence or absence of Sec1.

Conclusions:

• ExoFILT significantly enhances the efficiency and reliability of analyzing single-particle tracking data for exocytic events.
• The tool facilitates the discovery of novel mechanistic insights into the regulation of exocytosis.
• Automated analysis with ExoFILT opens avenues for large-scale studies on the molecular intricacies of exocytosis.