node2vec+

node2vec+ extends node2vec to perform biased random walks that incorporate edge weights to generate low-dimensional network embeddings for analysis of weighted biological networks.

Key Features:

• Unsupervised biased random-walk embedding: Uses biased random walks to embed networks into low-dimensional space.
• Weighted walk-bias computation: Modifies node2vec's bias calculations to account for edge weights and reduces to the original node2vec algorithm for unweighted graphs.
• Robustness to noise: Empirical evaluations on synthetic datasets show greater robustness to additive noise in weighted graphs compared to node2vec.
• Applications to genome-scale functional gene networks: Applied to genome-scale functional gene networks for gene function and disease prediction, outperforming node2vec in weighted scenarios.
• Comparison with graph neural networks: Outperforms graph neural networks such as GCN and GraphSAGE in gene classification tasks with limited training data.

Scientific Applications:

• Gene function prediction: Embeddings enable prediction of gene function from functional gene interaction networks.
• Disease association studies: Embeddings support disease association and disease prediction analyses on gene networks.
• Network-based biomarker discovery: Embeddings facilitate network-based biomarker discovery via downstream machine learning.

Methodology:

Extends node2vec by modifying biased random-walk bias calculations to incorporate edge weights, producing unsupervised low-dimensional embeddings via random-walk-based methods; evaluated on synthetic weighted graphs for additive-noise robustness and compared to node2vec, GCN, and GraphSAGE in gene classification tasks.