Epitome: predicting epigenetic events in novel cell types with multi-cell deep ensemble learning.

Area of Science:

• Computational biology
• Genomics
• Epigenetics

Background:

• Large epigenomics data consortiums offer opportunities to apply existing knowledge to new cell types and conditions.
• Predicting epigenomic signals like transcription factor binding is crucial for understanding cellular function.

Purpose of the Study:

• To develop a deep neural network model named Epitome.
• To leverage similarities in chromatin accessibility between reference and query cell types.
• To predict transcription factor binding and histone modification signals in novel cellular contexts.

Main Methods:

• Epitome, a deep neural network, learns chromatin accessibility similarities.
• It transfers transcription factor binding and histone modification signals from reference to query cell types based on profile similarity.
• The model was evaluated on its accuracy in predicting transcription factor binding sites.

Main Results:

• Epitome achieved state-of-the-art accuracy in predicting transcription factor binding sites on novel cellular contexts.
• The model's predictive performance improves with increased epigenetic signal data from both reference and query cell types.

Conclusions:

• Epitome effectively extrapolates epigenomic knowledge to new cell types.
• The deep learning approach enhances the prediction of transcription factor binding and histone modifications.
• This method holds potential for advancing epigenomic research with growing datasets.