DeepDualEnhancer: A Dual-Feature Input DNABert Based Deep Learning Method for Enhancer Recognition.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Enhancers are crucial cis-regulatory DNA elements controlling gene expression.
• Identifying enhancers is challenging due to their complex sequence features.

Purpose of the Study:

• To develop a robust deep learning model for accurate enhancer identification.
• To investigate the utility of genomic signals in enhancing enhancer recognition.

Main Methods:

• Proposed DeepDualEnhancer, a DNABert-based method utilizing multi-scale Convolutional Neural Network (CNN) and BiLSTM for DNA sequence analysis.
• Developed DeepDualEnhancer-genomic, incorporating genomic signals with transformer sequence attention for enhanced feature extraction.
• Utilized extensive datasets and rigorous validation including 5-fold cross-validation and independent testing.

Main Results:

• DeepDualEnhancer demonstrated superior performance compared to 20 other methods.
• The inclusion of genomic signals significantly improved enhancer recognition accuracy.
• Ablation studies confirmed the effectiveness of individual model components.

Conclusions:

• DeepDualEnhancer offers a powerful and accurate solution for enhancer identification.
• Integrating genomic signals is beneficial for improving enhancer recognition tasks.
• The findings advance the field of regulatory genomics and gene expression analysis.