GraphTGI: an attention-based graph embedding model for predicting TF-target gene interactions.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Transcription factor (TF)-target gene interactions are fundamental to transcriptional regulation.
• Current computational methods primarily focus on predicting TF binding sites, not direct interactions.
• Biological techniques for identifying TF-target interactions are limited.

Purpose of the Study:

• To develop a computational model for predicting TF-target gene interactions.
• To address the limitations of existing methods by directly predicting interactions.
• To leverage known interaction networks and gene features for improved prediction.

Main Methods:

• Formulated TF-target gene interaction prediction as a link prediction problem on a knowledge graph.
• Developed a deep learning model named GraphTGI.
• GraphTGI utilizes a graph attention-based encoder and a bilinear decoder.

Main Results:

• The GraphTGI model achieved an average AUC of 0.8864 +/- 0.0057 in 5-fold cross-validation.
• Demonstrated outstanding performance in predicting TF-target gene interactions on a real dataset.
• The model effectively learns patterns from known TF-target gene interactions.

Conclusions:

• GraphTGI is the first model to predict TF-target gene interactions by learning patterns from known networks.
• The model offers an effective and efficient approach for large-scale TF-target gene interaction prediction.
• The developed model advances the field of transcriptional regulation research.