Supervised enhancer prediction with epigenetic pattern recognition and targeted validation.

Area of Science:

• Genomics and Molecular Biology
• Computational Biology
• Epigenetics

Background:

• Enhancers are critical non-coding regulatory elements in the genome.
• Experimental characterization of enhancers has been historically challenging.
• Massively parallel assays have enabled large-scale enhancer characterization.

Purpose of the Study:

• To develop a computational framework for predicting enhancers using epigenetic features.
• To validate the model's accuracy and cross-species applicability.
• To differentiate enhancers from promoters based on transcription factor binding patterns.

Main Methods:

• Developed a framework using Drosophila STARR-seq (Stimulation, Assay for Transposon-based Element Accessibility and Reporter gene expression) with shape-matching filters.
• Integrated epigenetic features with supervised machine learning algorithms for enhancer prediction.
• Validated predictions using in vivo transgenic assays in mice and in vitro reporter assays in human cell lines.

Main Results:

• The developed model accurately predicts enhancers in Drosophila and can be transferred to predict mammalian enhancers without re-parameterization.
• Comprehensive validation confirmed the model's high accuracy across species (153 enhancers validated).
• Analysis of transcription factor binding patterns allowed for a secondary model to effectively distinguish enhancers from promoters.

Conclusions:

• The computational framework provides an accurate and versatile method for enhancer identification.
• The model's cross-species applicability and validation demonstrate its robustness.
• Transcription factor binding patterns serve as a key feature for distinguishing enhancers from promoters.