Prediction of protein subcellular localization in single cells

Area of Science:

• Computational Biology
• Proteomics
• Cell Biology

Background:

• Protein subcellular localization is critical for cellular function and disease.
• Current prediction models lack cell-type specificity and generalization to novel proteins.
• Existing datasets for protein localization are limited in scope.

Purpose of the Study:

• To develop a novel computational method for predicting protein subcellular localization.
• To enable prediction for unseen proteins and account for cell-type specificity.
• To address limitations in existing protein localization datasets and models.

Main Methods:

• Developed Prediction of Unseen Proteins' Subcellular localization (PUPS).
• Integrated a protein language model with an image inpainting model.
• Utilized both protein sequence and cellular images for prediction.

Main Results:

• Protein sequence input enabled generalization to unseen proteins.
• Cellular image input captured single-cell variability for cell-type-specific predictions.
• PUPS successfully predicted protein localization in new experiments beyond training data.

Conclusions:

• PUPS provides a robust framework for predicting differential protein localization.
• The method accurately predicts localization across cell lines and single cells.
• PUPS can identify changes in protein localization due to mutations.