DeepCORE: An interpretable multi-view deep neural network model to detect co-operative regulatory elements.

Area of Science:

• Genomics
• Computational Biology
• Epigenetics

Background:

• Gene transcription is crucial for cellular functions, traits, and diseases, regulated by complex networks of interacting elements.
• Understanding these co-operative regulatory elements (COREs) is key to deciphering gene expression control.

Purpose of the Study:

• To develop a novel deep learning method for identifying COREs by integrating genetic, epigenetic, and transcriptional data.
• To accurately predict gene expression and discover novel regulatory elements involved in transcription.

Main Methods:

• A multi-view attention-based deep neural network, DeepCORE, was developed to model relationships between different biological data types.
• DeepCORE utilizes an interpreter to extract attention values, map them to regulatory regions, and infer COREs based on correlated attention patterns.

Main Results:

• DeepCORE accurately predicted transcriptomes across various tissues and cell lines, outperforming existing state-of-the-art algorithms.
• Identified COREs were significantly enriched with known gene regulatory elements like promoters and enhancers.
• Novel regulatory elements discovered by DeepCORE exhibited epigenetic signatures consistent with histone modification patterns.

Conclusions:

• DeepCORE provides a powerful new approach for dissecting complex gene regulatory networks.
• The method successfully identifies known and novel regulatory elements, advancing our understanding of transcription control.