Precise measurement of nanoscopic septin ring structures with deep learning-assisted quantitative superresolution...

Area of Science:

• Cellular Biology
• Biophysics
• Microscopy and Imaging

Background:

• Superresolution microscopy and image analysis offer deep insights into macromolecular assemblies.
• Deep learning (DL) enhances automated data processing for high-resolution imaging.
• Current DL methods struggle with analyzing cellular structures that change size or shape, potentially introducing bias.

Purpose of the Study:

• To develop and validate DL models for accurate analysis of amorphous macromolecular complexes.
• To investigate bias sources in DL-based image recognition for superresolution microscopy.
• To establish a reliable method for measuring septin ring size and polymerization dynamics.

Main Methods:

• Utilized single molecule localization-based superresolution microscopy data of septin ring structures.
• Developed and rigorously tested DL models with real and simulated data across various structural phenotypes.
• Quantitatively compared model accuracy against the range of recognized structural variations.

Main Results:

• Identified a trade-off between measurement accuracy and the diversity of phenotypes recognized by DL models.
• Demonstrated that septin ring size is solely determined by the number of constituent subunits.
• Established a new experimental system for studying septin polymerization.

Conclusions:

• Validated DL models can overcome limitations in analyzing dynamic cellular structures.
• Septin ring assembly is directly proportional to subunit count, independent of complex shape.
• The developed methodology and experimental system advance the study of macromolecular complex dynamics.