FactorHNE

FactorHNE employs a factor graph-aggregated heterogeneous network embedding approach to generate node embeddings for predicting disease–gene associations and modeling multi-source heterogeneous biological relationships.

Key Features:

• Factor graph aggregation: Constructs factor graphs to aggregate diverse semantic relationships among heterogeneous biological entities.
• Multiple semantic factor graphs: Builds multiple semantic factor graphs to capture different semantic relationships inherent in multi-source data.
• Heterogeneous network embedding: Generates embeddings that represent heterogeneous nodes such as genes and diseases within a unified latent space.
• End-to-end multi-perspective loss function: Optimizes embeddings using an end-to-end multi-perspective loss function to improve representation quality.
• Node representation optimization: Produces high-quality node representations that support downstream predictive tasks.
• Disease–gene association prediction: Uses the learned embeddings specifically for predicting disease–gene associations.
• Integration of multi-source heterogeneous biological data: Aggregates relationships from multi-source heterogeneous biological data to inform embeddings.
• Scalability: Reported to scale to large biomedical network datasets.
• Interpretability: Provides interpretable embeddings and semantic factor structures to aid biological insight.
• Empirical performance: Reported to outperform existing models in performance and scalability in experimental evaluations.

Scientific Applications:

• Disease–gene association discovery: Prioritizes and predicts associations between diseases and genes using learned node embeddings.
• Modeling heterogeneous biological networks: Captures and analyzes semantic relationships among genes, diseases, and other heterogeneous entities.
• Understanding disease pathogenesis: Supports analysis of network relationships relevant to disease mechanisms.
• Therapeutic strategy development and large-scale analysis: Facilitates large-scale biomedical network analyses that can inform development of therapeutic strategies.

Methodology:

Constructs multiple semantic factor graphs from multi-source heterogeneous biological data, aggregates semantic relationships via a factor graph-aggregated heterogeneous network embedding approach, and optimizes embeddings using an end-to-end multi-perspective loss function for disease–gene association prediction.