Protein Condensate Atlas from predictive models of heteromolecular condensate composition

Area of Science:

• Cell Biology
• Biophysics
• Computational Biology

Background:

• Biomolecular condensates are crucial for cellular organization, but their diverse compositions and many undiscovered types present a challenge.
• Understanding protein localization drivers within condensates is key to deciphering their function.

Purpose of the Study:

• To develop a predictive methodology for identifying protein components of biomolecular condensates.
• To investigate the biophysical features governing protein localization into heteromolecular condensates.
• To create a comprehensive resource for condensate research.

Main Methods:

• Analysis of proteomics data from cellular condensates to identify key biophysical features.
• Development of a machine learning model linking protein sequence to condensate localization propensity.
• Proteome-wide application of the model and validation using immunohistochemical staining.
• Segmentation of the condensation-prone proteome into types based on interaction profiles to generate a Protein Condensate Atlas.

Main Results:

• Biophysical features like charge-mediated protein-RNA and hydrophobicity-mediated protein-protein interactions are critical for condensate localization, differing from homotypic phase separation drivers.
• The machine learning model successfully predicted numerous novel condensate-localizing proteins.
• The generated Protein Condensate Atlas contains clusters that align with known condensates and suggests potential new ones.

Conclusions:

• The developed methodology accurately predicts protein localization into biomolecular condensates.
• The Protein Condensate Atlas serves as a valuable tool for identifying components of known condensates and discovering novel ones.
• This work advances our understanding of cellular organization through the lens of condensate composition.