BraneMF

BraneMF integrates multilayer biological networks using a random walk-based matrix factorization to learn protein representations for functional analysis and prediction from gene co-expression, protein-protein interaction (PPI), genetic interaction, and metabolic networks.

Key Features:

• Multilayer Network Integration: Employs a random walk-based matrix factorization method to integrate information across multiple network layers.
• Node Representation Learning: Produces protein embeddings that reflect proximity and interactions across gene co-expression, PPI, genetic interaction, and metabolic networks.
• Application to Saccharomyces cerevisiae PPI data: Method has been tested on PPI networks of Saccharomyces cerevisiae to demonstrate applicability to real-world omics data.
• Performance Benchmarking: Benchmarked against state-of-the-art multilayer network integration methods across downstream tasks such as clustering, function prediction, and PPI prediction.
• Robustness Assessment: Evaluated via parameter sensitivity analysis to assess consistency and reliability of predictions.

Scientific Applications:

• Clustering: Facilitates clustering of proteins based on integrated network-derived representations to identify functionally related groups.
• Function Prediction: Enhances prediction of protein functions by leveraging comprehensive network-derived features.
• PPI Prediction: Improves accuracy of protein-protein interaction prediction using multilayer network embeddings.

Methodology:

Random walk-based matrix factorization applied to multilayer networks, with benchmarking against existing methods and parameter sensitivity analysis for robustness.