Heterogeneous graph embedding model for predicting interactions between TF and target gene.

Area of Science:

• Bioinformatics and Computational Biology
• Genomics and Molecular Biology
• Systems Biology

Background:

• Identifying transcription factor (TF)-target gene interactions is crucial for biomedical research.
• Experimental methods for TF-target gene identification are time-consuming, costly, and limited in scale.
• Existing computational methods often focus on TF binding sites rather than direct interactions.

Purpose of the Study:

• To develop a novel deep learning model, HGETGI, for predicting new TF-target gene interactions.
• To address the limitations of experimental and existing computational approaches for TF-target gene identification.
• To leverage known interactions and disease mechanisms for accurate prediction.

Main Methods:

• Proposed a deep learning prediction model named HGETGI.
• Model learns patterns from known TF-target gene interactions and their involvement in human disease mechanisms.
• Employs random walk for meta-path sampling and node embedding in a skip-gram manner for prediction.

Main Results:

• Achieved an average area under the curve (AUC) of 0.8519 ± 0.0731 in fivefold cross-validation on a real dataset.
• Case studies on NFKB1 and TP53 demonstrated high prediction accuracy, with 33 and 32 top-40 predictions confirmed, respectively.
• Validation against the hTftarget database confirmed the accuracy of predicted TF-target gene interactions.

Conclusions:

• The HGETGI method is effective and feasible for large-scale prediction of TF-target gene interactions.
• The model's ability to integrate interaction patterns and disease mechanisms enhances prediction accuracy.
• HGETGI offers a valuable computational tool for advancing biomedical research by facilitating TF-target gene discovery.