Bakta and Prokka call the same 87,859 CDS regions in 25 complete bacterial genomes, then give 51.7% of them different product names. Both wrap Prodigal, so they agree on gene boundaries and diverge on function. Grouped-by-genome ML shows the disagreement tracks database coverage, not sequence.
machine-learning random-forest scikit-learn bioinformatics-pipeline genome-annotation microbial-genomics comparative-genomics functional-annotation gene-prediction protein-function-prediction prodigal prokka prokaryotic-genomes bacterial-genomics bakta cds-prediction
-
Updated
Aug 23, 2026 - Python