Prediction of protein subcellular localization in single cells

Area of Science:

• Cell Biology
• Bioinformatics
• Computational Biology

Background:

• Protein subcellular localization is crucial for cellular function and disease pathogenesis.
• Existing protein localization datasets are limited in scope, covering only a fraction of human proteins and cell lines.
• There is a need for methods that can predict protein localization for novel proteins and cell types.

Purpose of the Study:

• To present a novel computational method, Predictions of Unseen Proteins' Subcellular localization (PUPS), for predicting protein subcellular localization.
• To enable generalization to proteins and cell lines not included in the training data.
• To capture cell-type-specific and single-cell variability in protein localization.

Main Methods:

• PUPS integrates a protein language model, utilizing protein sequences, with an image inpainting model, using cellular landmark images.
• The protein sequence component facilitates generalization to unseen proteins.
• The cellular image component enables cell-type-specific predictions and captures single-cell variability.

Main Results:

• PUPS successfully generalizes to predict protein localization in unseen proteins and cell lines.
• The method can assess variability in protein localization across different cell lines and within single cells.
• PUPS identified biological processes associated with proteins exhibiting variable localization.
• Experimental validation confirmed PUPS's ability to predict protein localization in new experiments beyond the training dataset.

Conclusions:

• PUPS offers a powerful tool for predicting protein subcellular localization, overcoming limitations of existing datasets.
• The method's ability to generalize and capture variability enhances our understanding of protein function and localization dynamics.
• PUPS facilitates the study of protein localization in novel contexts and disease-related investigations.