EssSubgraph

EssSubgraph predicts essential genes by integrating omics data with graph-structured network data using inductive representation learning and graph neural networks to enable cross-species essential gene identification.

Key Features:

• Inductive representation learning: Uses inductive representation learning to enable prediction on unseen nodes.
• Integration of network and omics: Combines graph-structured network data with omics features for joint representation.
• Graph neural networks: Trains graph neural networks on integrated representations to predict essential genes.
• Computational efficiency: Operates efficiently on large-scale networks without requiring learning entire gene interaction networks.
• Scalability: Scales to increasing network sizes while maintaining performance.
• Benchmarking on knockout datasets: Benchmarks performance using extensive lists of human essential genes derived from knockout datasets.
• Robustness to perturbation: Maintains stability of predictions under perturbations to network structure and gene features.
• Cross-species prediction: Demonstrates superior cross-species essential gene prediction performance compared to other methods.

Scientific Applications:

• Minimal genome analysis: Determining minimal genetic requirements of organisms by identifying essential genes.
• Disease gene identification: Pinpointing disease-associated genes through essential gene detection.
• Drug target discovery: Discovering potential drug targets among predicted essential genes.
• Gene function prediction: Predicting gene functions dynamically, including for unseen nodes.

Methodology:

Applies inductive representation learning to combine graph-structured network data with omics features, trains graph neural networks on these representations, and benchmarks predictions using lists of human essential genes from knockout datasets along with perturbation-based stability tests.