DeepICSH: a complex deep learning framework for identifying cell-specific silencers and their strength from the human...

Area of Science:

• Genomics
• Computational Biology
• Molecular Biology

Background:

• Silencers are noncoding DNA fragments that regulate gene expression, and their cell-specific variations are linked to gene expression and cancer.
• Current computational methods for silencer identification lack accuracy due to their reliance solely on DNA sequence, ignoring cell specificity.
• Identifying definitive biological signals for silencers remains challenging, hindering accurate computational identification.

Purpose of the Study:

• To develop a sophisticated deep learning framework, DeepICSH, for accurate identification of cell-specific silencers.
• To leverage multiple biological data sources and deep learning techniques to capture relevant signal combinations for silencer characterization.
• To introduce a deep learning framework for classifying strong and weak silencers using multi-omics data.

Main Methods:

• Developed DeepICSH, a deep learning framework utilizing deep convolutional neural networks, attention mechanisms, and skip connections.
• Integrated diverse biological signals and multi-omics data for comprehensive silencer feature extraction.
• Employed attention mechanisms for scoring and visualization of signal combinations and skip connections for feature fusion.

Main Results:

• DeepICSH demonstrated superior performance in silencer identification compared to state-of-the-art methods on HepG2 and K562 cell line data.
• Achieved favorable performance in classifying strong and weak silencers using a deep learning framework based on multi-omics data.
• The framework effectively captures biologically relevant signal combinations associated with silencers.

Conclusions:

• DeepICSH offers a promising advancement for the accurate identification and analysis of cell-specific silencers.
• The study highlights the potential of multi-omics data and deep learning in understanding gene regulation and complex diseases.
• DeepICSH provides a valuable tool for future research in genomics and cancer biology.