DNABERT2-CAMP: A Hybrid Transformer-CNN Model for E. coli Promoter Recognition.

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Accurate identification of promoter sequences in Escherichia coli is crucial for gene regulation and synthetic biology.
• Existing computational methods face challenges in modeling both long-range genomic dependencies and fine-grained local motifs, such as the degenerate -10 and -35 elements of σ70 promoters.

Purpose of the Study:

• To propose DNABERT2-CAMP, a hybrid deep learning framework for robust promoter identification.
• To integrate global contextual understanding with high-resolution local motif detection.

Main Methods:

• A balanced dataset of 8720 validated and negative 81-bp sequences was constructed from RegulonDB, literature, and the E. coli K-12 genome.
• The DNABERT2-CAMP model combines a pre-trained DNABERT-2 Transformer for global encoding with a custom Convolutional Neural Network-Attention-Mean Pooling (CAMP) module for local feature refinement.
• Performance was evaluated using 5-fold cross-validation and an independent external test set, reporting accuracy, ROC AUC, and Matthews correlation coefficient (MCC).

Main Results:

• DNABERT2-CAMP achieved 93.10% accuracy and 97.28% ROC AUC in cross-validation, outperforming existing methods like DNABERT.
• On an independent test set, the model demonstrated strong generalization with 89.83% accuracy and 92.79% ROC AUC.
• Interpretability analyses revealed biologically plausible attention over canonical promoter regions and CNN-identified motifs.

Conclusions:

• Synergistically combining pre-trained Transformers with convolutional motif detection significantly enhances promoter recognition accuracy and interpretability.
• The DNABERT2-CAMP framework provides a powerful and generalizable tool for genomic annotation and synthetic biology applications.