EP-DNN: A Deep Neural Network-Based Global Enhancer Prediction Algorithm.

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Enhancers are crucial regulatory elements controlling gene expression.
• Predicting enhancers accurately across diverse cell types remains challenging.
• Deep learning offers potential for complex pattern recognition in genomic data.

Purpose of the Study:

• To present EP-DNN, a novel deep neural network (DNN) protocol for enhancer prediction.
• To evaluate EP-DNN's performance in predicting enhancers using chromatin features in different cell types.
• To compare EP-DNN against existing state-of-the-art methods for enhancer prediction.

Main Methods:

• Utilized a deep neural network (DNN) architecture to extract enhancer signatures.
• Trained EP-DNN using p300 binding sites as enhancers and TSS/non-DHS sites as non-enhancers.
• Performed same-cell and cross-cell predictions, comparing results with DEEP-ENCODE and RFECS.

Main Results:

• EP-DNN achieved a superior validation rate of 91.6%, outperforming DEEP-ENCODE (85.3%) and RFECS (85.5%).
• EP-DNN demonstrated better scalability for predicting a larger number of enhancers.
• Cross-cell predictions (H1 → IMR90) were more accurate than same-cell predictions (IMR90 → IMR90).

Conclusions:

• EP-DNN effectively leverages deep learning expressivity for accurate enhancer prediction while avoiding overfitting.
• The protocol shows promise for cross-cell enhancer predictions, potentially reducing experimental costs.
• EP-DNN provides a robust foundation for exploring cell-type-specific regulatory element identification.