HGETGI

HGETGI predicts interactions between transcription factors and target genes using heterogeneous graph embedding and deep learning to model TF-gene regulatory relationships and disease associations.

Key Features:

• Deep Learning Approach: Employs a deep learning framework that leverages known TF-target gene interaction patterns and integrates disease mechanism information for prediction.
• Heterogeneous Graph Embedding: Represents transcription factors and genes as distinct node types and encodes diverse edge relationships within a heterogeneous graph.
• Meta-path Sampling with Random Walk: Applies random walk-based meta-path sampling to traverse the heterogeneous graph and generate node sequences for embedding.
• Node Embedding via Skip-gram Model: Learns dense vector representations of nodes from sampled sequences using a skip-gram model.

Scientific Applications:

• Large-scale TF-target interaction prediction: Enables genome-scale prediction of transcription factor–target gene interactions to inform studies of gene regulation and disease pathways.
• TF prioritization and validation: Supports prioritization and investigation of TFs such as NFKB1 and TP53, with predictions validated against external databases in case studies.

Methodology:

Integrates known TF-target interaction data with disease mechanism information; constructs a heterogeneous graph with TF and gene nodes and interaction/association edges; performs meta-path sampling via random walk to generate node sequences; learns node embeddings using a skip-gram model; evaluates predictive performance with fivefold cross-validation reporting an average AUC of 0.8519 ± 0.0731 and validates predictions through case studies on NFKB1 and TP53 confirmed by external databases.