Predicting enhancers in mammalian genomes using supervised hidden Markov models.

Area of Science:

• Genomics
• Epigenetics
• Gene Regulation

Background:

• Eukaryotic gene regulation involves complex interactions between enhancers and promoters.
• While promoter elements are well-annotated, enhancer characteristics and locations remain elusive.
• High-throughput methods generate epigenetic data, but existing prediction tools often neglect enhancer-specific biological knowledge.

Purpose of the Study:

• To develop a novel computational method for predicting active enhancers.
• To incorporate prior biological knowledge of enhancer molecular structure into prediction models.
• To improve the accuracy and resolution of enhancer identification.

Main Methods:

• Developed enhancer Hidden Markov Model (eHMM), a supervised machine learning approach.
• eHMM models enhancers and promoters as a central accessible DNA region flanked by nucleosomes with specific histone modifications.
• Utilized chromatin accessibility and ChIP-seq data for histone modifications.

Main Results:

• eHMM successfully predicts enhancers with high precision and recall.
• Performance was evaluated across different cell types and developmental stages.
• eHMM consistently outperformed state-of-the-art methods in accuracy and resolution.

Conclusions:

• eHMM predicts active enhancers using chromatin accessibility and minimal histone modification data.
• The model's parameters are interpretable, unlike 'black box' methods.
• eHMM serves as a stand-alone tool for enhancer prediction, providing precise targets for further experimental validation.