EnContact: predicting enhancer-enhancer contacts using sequence-based deep learning model.

Area of Science:

• Genomics
• Computational Biology
• Molecular Biology

Background:

• Chromatin contacts between regulatory elements are vital for gene transcription and disease mechanisms.
• Current computational tools primarily predict enhancer-promoter interactions, neglecting enhancer-enhancer (E-E) contacts.

Purpose of the Study:

• To introduce EnContact, a novel deep learning model for predicting E-E contacts using genomic sequences.
• To evaluate EnContact's predictive performance and compare it against existing methods.

Main Methods:

• Developed a deep learning model (EnContact) that utilizes genomic sequences for E-E contact prediction.
• Validated the model using HiChIP data from seven cell lines.
• Compared EnContact's performance against baseline computational methods.

Main Results:

• EnContact demonstrated statistically significant predictive ability for E-E contacts.
• The model outperformed existing baseline methods in predicting E-E interactions.
• EnContact identified finer-mapping E-E interactions and a class of active 'hub enhancers' across cell lines.

Conclusions:

• EnContact effectively predicts E-E interactions by learning features directly from genomic sequences.
• The model enhances the exploration of E-E contacts, contributing to a deeper understanding of transcriptional regulation and disease.
• Identified hub enhancers suggest their broad regulatory roles across different cell types.