PIFiA

PIFiA annotates protein function from single-cell fluorescence microscopy images by learning self-supervised image representations to capture protein localization patterns.

Key Features:

• Self-supervised representation learning: Learns image representations without supervised labels using a self-supervised learning methodology.
• Single-cell fluorescence microscopy: Operates on fluorescence microscopy micrographs at single-cell resolution.
• Protein feature profiles: Generates quantitative protein feature profiles from images of fluorescently tagged proteins.
• Yeast ORF-GFP dataset: Was developed and tested using the global yeast ORF-GFP collection of fluorescently tagged proteins.
• Hierarchical feature clustering: Clusters extracted features into a hierarchical structure that reflects functional organization within cells.
• Improved molecular representations: Demonstrates superior molecular representation learning performance relative to existing approaches.
• Multi-localizing protein detection: Identifies proteins that localize to multiple cellular compartments.
• Cell population heterogeneity analysis: Enables exploration of heterogeneity across cell populations based on protein localization features.
• Downstream analytical tasks: Supports downstream analyses to delineate functional modules and generate hypotheses for experimental validation.

Scientific Applications:

• Protein functional annotation: Annotates protein function and localization from microscopy images at single-cell resolution.
• Localization pattern discovery: Detects and characterizes protein localization patterns and multi-localization behavior.
• Functional module delineation: Identifies clusters of proteins that reflect cellular functional organization.
• Single-cell heterogeneity studies: Quantifies and compares protein localization variability across individual cells.
• Feature-driven hypothesis generation: Produces feature profiles used to prioritize candidates for experimental follow-up such as colocalization assays.
• Benchmarking representation methods: Serves as a comparative approach for molecular representation learning from imaging data.

Methodology:

Applies self-supervised representation learning to single-cell fluorescence microscopy images to extract protein feature profiles and clusters those features into a hierarchical structure for downstream analyses.