GNNHap

GNNHap identifies causative genetic factors in mouse genetic models by applying a graph neural network to integrated genetic, literature, and biological network data.

Key Features:

• Automated analysis pipeline: Streamlines analysis of mouse genetic model data to prioritize candidate causal variants and genes.
• Integration of diverse data sources: Assesses allelic associations based on strain response patterns and integrates literature-derived evidence from a 29 million published-paper repository to evaluate gene-phenotype relationships.
• Incorporation of biological networks: Leverages protein-protein interaction networks and protein sequence features to add biological context to candidate gene evaluation.
• Advanced neural network architecture: Employs a graph neural network model and reports superior predictive performance compared to traditional linear neural networks.
• GWAS false-positive mitigation: Addresses false positive associations commonly encountered in genome-wide association studies (GWAS) through integrated, multi-source analysis.

Scientific Applications:

• Diabetes/Obesity: Identifies novel causative genetic factors associated with diabetes and obesity traits in murine models.
• Cataract formation: Detects candidate genes implicated in cataract development in mice.
• Gene knockout validation: Supports validation of predicted causal factors via phenotypic observations in previously analyzed gene knockout mice.

Methodology:

Integrates allelic associations (strain response patterns), gene-phenotype relationships, literature-derived signals from a 29 million-paper repository, protein-protein interaction networks, and protein sequence features into a graph representation, then applies a graph neural network to predict causal genetic factors and compares performance to linear neural network models.