SEMORE: SEgmentation and MORphological fingErprinting by machine learning automates super-resolution data analysis.

Area of Science:

• Biophysics
• Cell Biology
• Computational Biology

Background:

• Protein assembly morphology influences cellular functions, both normal and disease-related.
• Super-resolution microscopy offers high resolution for studying these assemblies, but lacks universal analytical tools.
• Existing methods struggle to universally extract and quantify complex protein structures from microscopy data.

Purpose of the Study:

• To develop a universal, semi-automatic machine learning framework for analyzing super-resolution microscopy data of protein assemblies.
• To enable robust quantification of protein assembly morphology and temporal evolution.
• To provide a general platform for dissecting and characterizing biological structures in super-resolution imaging.

Main Methods:

• Implemented a multi-layered density-based clustering module for dissecting biological assemblies.
• Developed a morphology fingerprinting module using geometric and kinetics-based descriptors for quantification.
• Applied the SEMORE framework to diverse super-resolution datasets, including simulations and experimental data (insulin aggregates, NPCs, etc.).

Main Results:

• SEMORE successfully extracts and quantifies various protein assemblies and their morphology evolution from diverse super-resolution data.
• The framework provides quantitative insights, such as classifying insulin aggregation pathways and determining NPC geometry.
• Analysis is significantly accelerated, with results obtained in minutes.

Conclusions:

• SEMORE is a versatile and efficient machine learning framework for analyzing super-resolution microscopy data of protein assemblies.
• It addresses the need for universal analytical tools, enabling detailed quantification of structure and dynamics.
• The time-aware nature of SEMORE supports the analysis of 4D super-resolution data, advancing the field.