Balrog: A universal protein model for prokaryotic gene prediction

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• The rapid increase in sequenced microbial genomes necessitates efficient annotation tools.
• Existing gene finding methods often require genome-specific retraining, limiting scalability.

Purpose of the Study:

• To develop a universal gene finding model for prokaryotes.
• To create a scalable and accurate gene annotation system.

Main Methods:

• A temporal convolutional network was trained on amino acid sequences from a diverse set of microbial genomes.
• The trained model was integrated into the Balrog (Bacterial Annotation by Learned Representation Of Genes) gene finding system.

Main Results:

• The Balrog system demonstrates performance matching or exceeding current state-of-the-art gene finding tools.
• Balrog does not require genome-specific training, offering a universal solution.

Conclusions:

• The developed universal model and Balrog system provide a significant advancement in microbial genome annotation.
• Balrog offers a scalable, accurate, and freely available tool for the research community.