GraphLoc: a graph neural network model for predicting protein subcellular localization from immunohistochemistry...

Area of Science:

• Biochemistry
• Computational Biology
• Bioinformatics

Background:

• Protein subcellular localization is crucial for understanding protein function and disease.
• Immunohistochemical (IHC) images are vital for visualizing protein distribution in tissues.
• Existing image-based prediction methods struggle with multi-label proteins and pattern variations.

Purpose of the Study:

• To develop an advanced model for accurate protein subcellular location prediction using IHC images.
• To identify novel location biomarker proteins and protein network members in cancer tissues.

Main Methods:

• Proposed GraphLoc, a multi-label, multi-instance deep graph convolutional neural network model.
• Constructed graphs from multiple IHC images per protein for learning.
• Employed graph convolutions for protein-level representation learning.
• Utilized a dynamic threshold method for multi-label prediction.

Main Results:

• GraphLoc demonstrated promising performance in image-based protein subcellular location prediction.
• The model offers interpretability for its predictions.
• Applied GraphLoc to identify candidate location biomarkers and protein network members.
• A significant portion of predictions were supported by existing literature, with new candidates proposed.

Conclusions:

• GraphLoc is an effective tool for protein subcellular location prediction from IHC images.
• The model aids in biomarker discovery and understanding protein interaction networks.
• Predicted candidates offer valuable directions for future experimental validation.