supervised-learning

supervised-learning applies machine learning to molecular interaction networks to predict and assign human genes to specific functions, diseases, and traits.

Key Features:

• Network-Based Approach: Leverages the full connectivity of genes within molecular interaction networks to capture local network properties.
• Comprehensive Benchmarking: Evaluated across hundreds of diverse prediction tasks and multiple networks using stringent evaluation schemes.
• Performance Superiority: Outperforms label-propagation by utilizing full network connectivity rather than relying solely on network topology.
• Comparison with Node-Embeddings: Surpasses node-embedding methods such as node2vec that aim to concisely represent network connectivity.

Scientific Applications:

• Gene Prioritization: Prioritizes genes associated with functions, diseases, and traits for downstream experimental or computational follow-up.
• Pathway and Phenotype Analysis: Facilitates investigation of pathways and phenotypes by predicting gene attributes relevant to biological processes.
• Disease–Gene Association Discovery: Aids discovery of disease–gene associations through high-accuracy prediction of gene–disease links.

Methodology:

Trains machine learning models on annotated datasets within molecular interaction networks, emphasizes full network connectivity, benchmarks performance across hundreds of prediction tasks and multiple networks using stringent evaluation schemes, and compares results to label-propagation and node-embedding approaches (e.g., node2vec).