A Spectral Clustering-Based Approach for Balancing Data in TF-Target Gene Interaction Prediction Using Heterogeneous...

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Identifying transcription factor (TF) target gene interactions is vital for understanding gene regulation, biological processes, and diseases.
• Current prediction methods face challenges due to high experimental costs, biological complexity, and significant data imbalance, limiting their performance.

Purpose of the Study:

• To develop an effective computational method for predicting TF-target gene interactions that addresses data imbalance.
• To improve the accuracy and robustness of TF-target gene interaction prediction for molecular biology and precision medicine.

Main Methods:

• A novel approach integrating sample selection with spectral clustering to balance the dataset of known and unknown TF-target interactions.
• Construction of an adjacency matrix from known interactions, followed by spectral clustering to partition data and select diverse unknown interactions.
• Application of a deep learning model utilizing random walk sampling and skip-gram embeddings for learning biological network representations.

Main Results:

• The proposed method achieved an average Area Under the Curve (AUC) of 0.9575 ± 0.0044 in five-fold cross-validation.
• Demonstrated superior performance compared to existing methods in predicting TF-target gene interactions.
• Successfully addressed the challenge of data imbalance, leading to improved prediction accuracy.

Conclusions:

• The developed approach offers a robust framework for discovering novel TF-target gene interactions.
• This method enhances prediction accuracy and effectively resolves data imbalance issues in TF-target gene interaction prediction.
• Provides valuable insights for molecular biology research and the advancement of precision medicine.