PromoterAtlas: decoding regulatory sequences across Gammaproteobacteria using a transformer model

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Deep learning, especially transformer architectures, has advanced biological sequence analysis.
• Bacterial promoter prediction models are limited by small datasets, species-specific training, and binary classification.

Purpose of the Study:

• To develop a comprehensive bacterial promoter annotation framework using a novel deep learning model.
• To improve the prediction and understanding of bacterial regulatory elements across diverse species.

Main Methods:

• Developed PromoterAtlas, a 1.8M parameter transformer model trained on 9M regulatory sequences from 3371 gammaproteobacterial species.
• Utilized the model to create a whole-genome promoter annotation tool for Gammaproteobacteria.
• Analyzed model embeddings to understand cross-species evolutionary relationships and regulatory sequence information encoding.

Main Results:

• The model accurately recognizes diverse regulatory elements, including ribosomal binding sites, promoters, transcription factor binding sites, and terminators.
• Promoter predictions are validated and associated with different sigma factors.
• Model embeddings reveal evolutionary relationships, clustering promoters by sigma factor identity, not species.
• Embeddings encode regulatory information enabling prediction of transcription and translation levels.

Conclusions:

• PromoterAtlas provides a powerful tool for bacterial promoter annotation and analysis.
• The model's insights into regulatory sequence evolution and function have implications for bacterial biology and synthetic biology.
• PromoterAtlas facilitates the engineering of bacterial regulatory sequences.