CT-TADB predicts TAD boundaries without Hi-C by integrating DNA sequences and epigenomic features

Area of Science:

• Genomics and Bioinformatics
• Computational Biology
• Epigenetics

Background:

• Topologically associating domains (TADs) are key to 3D genome organization, influencing gene regulation and stability.
• Predicting TAD boundaries computationally is difficult due to complex DNA sequence and epigenomic interactions.
• Existing methods often rely on Hi-C data, limiting accessibility.

Purpose of the Study:

• To develop a novel computational framework, CT-TADB, for predicting TAD boundaries.
• To integrate DNA sequence with histone modification and CTCF binding signals for enhanced prediction accuracy.
• To predict TAD boundaries without requiring Hi-C data.

Main Methods:

• Developed CT-TADB, a hybrid Convolutional Neural Network (CNN)-Transformer model.
• Integrated DNA sequence, histone modification, and CTCF binding data.
• Trained and validated the model on six human cell lines and cross-species datasets.

Main Results:

• CT-TADB achieved high prediction accuracy (AUC 0.932-0.950) in human cell lines.
• Demonstrated robust performance on independent datasets and strong cross-species generalization (human-to-mouse AUC > 0.80).
• Identified CTCF as a dominant boundary factor and revealed long-range CTCF dependencies. Identified a clinically relevant PITX2-associated TAD boundary.

Conclusions:

• CT-TADB offers a practical and accurate method for predicting TAD boundaries using accessible epigenomic data.
• The framework enhances the study of chromatin architecture and its role in gene regulation and disease.
• Provides a complementary approach to Hi-C for investigating 3D genome organization.