GraphTGI

GraphTGI predicts transcription factor (TF)-target gene interactions to infer transcriptional regulatory relationships from known interaction networks.

Key Features:

• Graph attention-based autoencoder model: Utilizes a graph attention mechanism within an autoencoder framework to model complex patterns in TF-target gene interaction networks.
• Integration of sequential and chemical gene characteristics: Incorporates both sequence-derived and chemical features of genes to enhance interaction prediction accuracy.
• Link prediction on a knowledge graph: Formulates TF-target interaction inference as a link prediction problem on a knowledge graph representing known interactions.
• Deep learning architecture: Implements a graph attention-based encoder paired with a bilinear decoder to capture and reconstruct potential interactions.
• Performance evaluation: Evaluated with 5-fold cross-validation yielding an average Area Under the Curve (AUC) of 0.8864 ± 0.0057.

Scientific Applications:

• Prediction of TF-target gene interactions: Enables identification of putative regulatory links between transcription factors and target genes.
• Transcriptional regulation analysis: Supports studies of transcriptional regulatory mechanisms by providing candidate regulatory interactions.
• Large-scale network inference: Facilitates genome-scale inference of regulatory networks and discovery of novel regulatory relationships.

Methodology:

Employs a graph attention-based autoencoder with a graph attention encoder and bilinear decoder, integrates sequential and chemical gene features, formulates prediction as link prediction on a knowledge graph, and assesses performance using 5-fold cross-validation with AUC.