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

The morphology of protein assemblies impacts their behavior and contributes to beneficial and aberrant cellular responses. While single-molecule localization microscopy (SMLM) provides the required spatial resolution, the lack of universal robust analytical tools limits this technique. This paper presents SEMORE, a semi-automatic machine learning framework for universal analysis of super-resolution data. SEMORE uses a multi-layered density-based clustering module and a morphology fingerprinting module for quantification. Its application to simulations and diverse SMLM data (time-resolved insulin aggregates, nuclear pore complexes, fibroblast growth receptor 1, Syntaxin 1a, and ryanodine receptors) demonstrates its ability to extract and quantify protein assemblies, their temporal morphology evolution, and provide quantitative insights.

Steen W. B. Bender, Marcus W. Dreisler, Min Zhang, Jacob Kæstel-Hansen, Nikos S. Hatzakis