DEEP: a general computational framework for predicting enhancers.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Transcription regulation in multicellular eukaryotes involves DNA functional elements like enhancers, which can be located distally to genes.
• Identifying these distal regulatory elements poses a significant bioinformatics challenge.
• Existing computational enhancer prediction methods face issues with performance inconsistency, class imbalance, and ad hoc candidate selection.

Purpose of the Study:

• To develop a novel ensemble prediction framework, named DEEP, for accurate identification of DNA enhancers.
• To address the limitations of existing computational models in enhancer prediction.
• To streamline the analysis of enhancer properties across diverse cellular conditions.

Main Methods:

• DEEP is an ensemble prediction framework integrating three diverse components.
• It trains multiple individual classification models that are combined to classify DNA regions as enhancers or non-enhancers.
• Features used include histone modification marks and sequence characteristics.

Main Results:

• DEEP demonstrated superior performance compared to four state-of-the-art methods on ENCODE data.
• On FANTOM5 data, DEEP achieved 90.2% accuracy and 90% geometric mean (GM) across 36 tissues.
• DEEP-VISTA, tested on VISTA database data, achieved 80.1% GM and 89.64% accuracy on an independent set.

Conclusions:

• DEEP offers a robust and accurate computational framework for enhancer prediction.
• The framework shows high performance across various datasets and cellular conditions.
• DEEP provides a valuable tool for understanding gene regulatory mechanisms and is publicly available.