Annotation and automated segmentation of single-molecule localisation microscopy data.

Area of Science:

• Cell biology
• Microscopy
• Biophysics

Background:

• Single Molecule Localisation Microscopy (SMLM) generates high-resolution molecular localization data as coordinate tables.
• Analyzing SMLM data often involves visualizing molecular distributions and performing cluster analysis.
• Segmentation of SMLM data is crucial for isolating localizations within specific cellular regions before downstream analysis.

Purpose of the Study:

• To develop and evaluate a computational pipeline for annotating SMLM data, focusing on accurate membrane and cell segmentation.
• To compare different segmentation methods, including thresholding and machine learning, for SMLM datasets.
• To enable precise downstream analysis of molecular distributions within distinct cellular compartments.

Main Methods:

• Development of an SMLM data annotation pipeline.
• Comparison of membrane segmentation approaches: Otsu thresholding and machine learning models (specifically Cellpose).
• Application of the pipeline to dSTORM images of cell pellets stained for EGFR and EREG.

Main Results:

• A retrained Cellpose model demonstrated superior performance in membrane segmentation compared to other tested methods.
• The pipeline successfully segmented SMLM data, allowing differentiation between membrane and intracellular localizations.
• The optimized segmentation facilitated downstream cluster analysis of molecular distributions.

Conclusions:

• The developed pipeline enhances the analysis of SMLM data by providing robust segmentation of cellular structures.
• Retrained machine learning models, like Cellpose, are highly effective for membrane segmentation in SMLM imaging.
• This approach is broadly applicable for advanced SMLM data analysis, particularly for studying protein localization and interactions.