Identification of Protein Subcellular Localization With Network and Functional Embeddings

Area of Science:

• Computational biology
• Bioinformatics
• Molecular cell biology

Background:

• Protein function is intrinsically linked to its location within a cell.
• Accurate prediction of subcellular localization is crucial for understanding protein function.
• Existing computational methods for subcellular localization prediction need enhancement, particularly in protein representation.

Purpose of the Study:

• To develop an improved computational method for predicting protein subcellular localization.
• To leverage functional and network information for more accurate protein representations.
• To enhance the classification model for subcellular localization prediction.

Main Methods:

• Learning functional embeddings from KEGG/GO terms for protein representation.
• Characterizing network embeddings of proteins using protein-protein interaction networks.
• Combining functional and network embeddings to create novel protein representations.
• Constructing a classification model using these integrated embeddings.

Main Results:

• A benchmark dataset comprising 4,861 proteins across 16 locations was utilized.
• The developed embedding-based method achieved a Matthews correlation coefficient of 0.872.
• The proposed method demonstrated superior performance compared to several conventional prediction techniques.

Conclusions:

• The integration of functional and network embeddings offers a powerful new strategy for protein subcellular localization prediction.
• The developed method provides a significant advancement over existing computational approaches.
• This approach holds promise for advancing our understanding of protein function and cellular processes.