DARDN: A Deep-Learning Approach for CTCF Binding Sequence Classification and Oncogenic Regulatory Feature Discovery.

Area of Science:

• Genomics
• Cancer Biology
• Computational Biology

Background:

• Gene regulation in cancer is crucial for understanding disease mechanisms.
• CCCTC-binding factor (CTCF) plays a role in cancer-specific gene transcription.
• Identifying sequence features of CTCF binding sites can reveal cancer-driving factors.

Purpose of the Study:

• To develop a computational method for predicting cancer-specific CTCF binding sites from long DNA sequences.
• To identify DNA sequence features associated with cancer-specific CTCF binding.
• To discover potential oncogenic transcription factors in various cancer types.

Main Methods:

• Utilized convolutional neural networks (CNNs) for sequence prediction.
• Employed DeepLIFT for model interpretability and feature attribution.
• Applied the method to CTCF binding sites in T-cell acute lymphoblastic leukemia (T-ALL) and other cancers.

Main Results:

• DNAResDualNet (DARDN) accurately classifies cancer-specific CTCF binding sites.
• Identified sequence motifs linked to transcription factors active in specific cancers.
• Discovered potential oncogenic transcription factors in T-ALL, AML, BRCA, CRC, LUAD, and PRAD.

Conclusions:

• Advanced machine learning, like DARDN, is powerful for discovering biologically meaningful patterns in complex genomic data.
• This approach aids in identifying novel therapeutic targets by uncovering cancer-specific regulatory mechanisms.
• The study highlights the utility of deep learning for feature discovery in high-throughput sequencing data.