node2loc

node2loc predicts protein subcellular locations by learning node2vec graph embeddings from a protein-protein interaction network and classifying these embeddings with a recurrent neural network to infer protein localization.

Key Features:

• Distributed Representation Learning: Uses node2vec to generate graph embeddings of proteins from a protein-protein interaction (PPI) network.
• Recurrent Neural Network (RNN) Integration: Processes the learned embeddings with a Recurrent Neural Network to predict protein subcellular locations.
• Handling Class Imbalance with SMOTE: Applies the Synthetic Minority Over-sampling Technique (SMOTE) to augment underrepresented subcellular location classes.
• Performance Reporting: Reports classification performance using the Matthews correlation coefficient (MCC), with an achieved MCC of 0.812.

Scientific Applications:

• Protein subcellular localization prediction: Predicts protein subcellular locations by integrating protein-protein interaction data and learned embeddings.
• Functional and contextual inference: Supports interpretation of protein function and cellular context when experimental localization data are unavailable or difficult to obtain.

Methodology:

Generates node2vec embeddings from a protein-protein interaction network, applies a Recurrent Neural Network for classification, uses SMOTE to address class imbalance, constructs a benchmark dataset comprising 16 distinct subcellular locations, and evaluates performance using the Matthews correlation coefficient (MCC = 0.812).