Identifying Protein Subcellular Locations With Embeddings-Based node2loc.

Area of Science:

• Bioinformatics
• Computational Biology
• Molecular Biology

Background:

• Protein subcellular localization is crucial for understanding protein function.
• Protein-protein interactions (PPIs) can indicate shared subcellular locations.
• Accurate localization prediction requires integrating PPI data.

Purpose of the Study:

• To introduce node2loc, a novel network embedding-based method for predicting protein subcellular locations.
• To leverage protein-protein interaction networks for improved localization prediction.
• To address the challenge of class imbalance in subcellular location datasets.

Main Methods:

• Utilized node2vec to learn protein embeddings from PPI networks.
• Employed a recurrent neural network (RNN) for classification using learned embeddings.
• Applied Synthetic Minority Over-sampling Technique (SMOTE) to handle imbalanced location data.

Main Results:

• Achieved a Matthews correlation coefficient (MCC) of 0.800 on a human benchmark dataset with 16 locations.
• Demonstrated superior performance compared to baseline methods.
• Showed improved results on a Yeast benchmark dataset with 17 locations.

Conclusions:

• Network embeddings from PPIs possess discriminative power for protein subcellular location prediction.
• node2loc offers a promising approach for enhancing localization prediction accuracy.
• Limitations include its transductive nature and reliance on annotated PPI networks.