EmbedTAD Using Graph Embedding and Unsupervised Learning to Identify TADs from High-Resolution Hi-C Data.

Area of Science:

• Genomics
• Computational Biology
• Epigenetics

Background:

• Topologically Associating Domains (TADs) are crucial for genome organization and function.
• Identifying TADs is essential for studying immune system dynamics and chromosomal structure.
• Existing methods for TAD identification can be computationally intensive.

Purpose of the Study:

• To introduce EmbedTAD, a novel computational method for identifying TADs.
• To leverage graph embedding and clustering for efficient TAD detection from high-resolution Hi-C data.
• To validate EmbedTAD's performance in detecting TAD rearrangements and boundary protein enrichment.

Main Methods:

• Utilized NetMF, a low-resource graph embedding technique.
• Employed HDBSCAN for clustering embeddings to define TAD regions.
• Applied EmbedTAD to high-resolution Hi-C data, including during T-cell differentiation.

Main Results:

• EmbedTAD successfully identified TAD rearrangements during T-cell differentiation.
• The method differentiated between active and inactive cells based on TAD structures.
• EmbedTAD demonstrated reproducibility by recovering TADs found in PLAC-seq data.
• Detected distinct ChIP-seq signals (CTCF, RAD21, SMC3) at identified TAD boundaries.

Conclusions:

• EmbedTAD provides a reliable and computationally efficient method for TAD identification.
• The tool outperforms existing state-of-the-art methods in TAD detection.
• EmbedTAD facilitates the study of genome organization and its relation to cellular processes.