Speaker
Description
Gene calling is a critical step in genome annotation, where errors propagate into downstream biological interpretation. We present a hybrid approach that integrates machine learning into BBTools CallGenes by adding a neural advisory score to candidate open reading frames (ORFs) before dynamic-programming selection. Rather than replacing the existing gene-calling framework, the model complements it by improving ORF score discrimination while preserving the speed and structure of the original pipeline. Using curated NCBI-derived training and evaluation datasets, this neural augmentation improves overall calling accuracy, reducing false positives and false negatives relative to baseline CallGenes. These results show that lightweight neural rescoring can deliver measurable accuracy gains in practical, high-throughput prokaryotic gene annotation workflows.
