Characterizing Promoter and Enhancer Sequences by a Deep Learning Method.

Area of Science:

• Genomics
• Molecular Biology
• Bioinformatics

Background:

• Promoters and enhancers are key gene regulatory elements.
• Both elements are bidirectionally transcribed, producing sense and/or antisense RNA.
• Distinct transcriptional mechanisms for promoters versus enhancers remain unclear.

Purpose of the Study:

• To differentiate promoter and enhancer regulatory elements using sequence features.
• To investigate the role of transcription start site (TSS) architecture in regulatory element function.
• To develop a deep learning model for analyzing regulatory element sequences.

Main Methods:

• Utilized a deep learning (DL) approach with a convolutional neural network (CNN).
• Employed the saliency algorithm to interpret CNN predictions.
• Analyzed high-throughput sequencing data of regulatory elements.

Main Results:

• The CNN model achieved high predictive performance, outperforming other classifiers.
• Identified significant sequence differences between enhancers and promoters.
• Discovered that regions 20-120 bp downstream of bidirectional TSSs impact RNA stability.
• Promoter-associated regions exhibit higher GC content than enhancer regions, aiding classification.

Conclusions:

• Deep learning effectively captures complex TSS architectures.
• Genomic regions surrounding TSSs are crucial for RNA stability.
• GC-biased characteristics near TSSs are critical determinants distinguishing promoter activity.