HTAD: a human-in-the-loop framework for supervised chromatin domain detection.

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Topologically Associating Domains (TADs) are fundamental to genome architecture, impacting gene regulation and disease.
• Accurate TAD identification is crucial but challenging due to complex genomic structures and nested domains.

Purpose of the Study:

• To introduce HTAD, a novel human-in-the-loop computational tool for high-accuracy TAD identification.
• To overcome limitations of existing TAD callers in detecting complex and hierarchical genomic structures.

Main Methods:

• HTAD employs machine learning combined with active learning for interactive human supervision.
• Feature extraction for potential TAD border pairs is followed by an iterative labeling process.

Main Results:

• HTAD demonstrates superior performance compared to state-of-the-art methods on public and synthetic datasets.
• The study reveals highly hierarchical TAD structures, offering new insights into genome organization.

Conclusions:

• HTAD provides an accurate and effective human-in-the-loop solution for identifying complex TADs.
• This approach enhances our understanding of genome architecture, transcriptional regulation, and disease mechanisms.