Abstract
Shotgun metagenomics enables taxonomic analysis of microbial communities by aligning sequencing reads to reference genomes, for which interpretation of alignment results often lacks standardization and relies on arbitrary abundance thresholds. This can bias species detection, especially for low-abundance or taxonomically complex genera like Bacillus, where closely related species may differ in safety and function, and their co-occurrence increases misclassification risk. This study presents a bioinformatics framework for defining detection thresholds of biological contaminations in samples using nanopore shotgun metagenomics data, demonstrated through a case study on Bacillus subtilis sensu lato (s.l.) and Bacillus cereus s.l. contaminations in food enzyme (FE) products. The framework was developed by employing in silico mixes of isolate sequencing data of different B. subtilis and B. cereus species, and uses the tool KMA for taxonomic classification with post-processing steps based on template identity to differentiate true positives from false positives, coupled with curation of the underlying reference genomic database. The performance of the developed framework was afterwards validated with five in vitro mixes mimicking potential FE contaminations. Finally, the applicability of the validated framework was evaluated with six real and well-characterized commercial contaminated FE samples, confirming its ability to accurately detect B. subtilis and B. cereus contaminants, even at low abundances up to a relative abundance of 1 %. In conclusion, we present a bioinformatics framework allowing reliable species-level detection of challenging low-level contaminants in samples using nanopore shotgun metagenomics sequencing, which was successfully applied to identify B. subtilis and B. cereus contaminations in FE products.