PIFiA: Self-supervised Approach for Protein Functional Annotation from Single-Cell Imaging Data.

Area of Science:

• Cell biology
• Proteomics
• Bioimage analysis

Background:

• Fluorescence microscopy provides single-cell resolution of protein localization, crucial for understanding cellular functions.
• Extracting meaningful biological information from cell images is challenging due to noise and the need for supervised labels.
• Existing methods struggle with robust, noise-invariant feature learning and often require extensive manual annotation.

Approach:

• Developed PIFiA (Protein Image-based Functional Annotation), a novel self-supervised approach for protein functional annotation.
• Applied PIFiA to the global yeast ORF-GFP collection, generating protein feature profiles from single-cell images.
• PIFiA learns robust and noise-invariant features without relying on supervised labels.

Key Points:

• PIFiA outperforms existing methods in molecular representation learning from imaging data.
• Extracted features enable hierarchical functional organization, cell population heterogeneity studies, and identification of multi-localizing proteins.
• Downstream analyses revealed previously unappreciated biological roles for several proteins, confirmed via colocalization assays.

Conclusions:

• PIFiA offers a powerful, self-supervised framework for quantitative analysis of protein organization in cells.
• The approach facilitates the discovery of novel protein functions and cellular mechanisms.
• PIFiA, accessible via an interactive website, serves as a valuable resource for the research community.