In silico prediction of Y. enterocolitica O:3, O:5/5,27, O:8 and O:9 serotypes and biotypes from whole-genome sequencing data

Yersinia enterocolitica is a significant food-borne pathogen with diverse O-antigen related serotypes, among which O:3, O:5,27, O:8, and O:9 are the most frequently implicated in human disease. Conventional serotyping methods, such as phenotypic assays and targeted PCR, are labor-intensive and lack scalability. This paper presents YersiniaSerotyper, an in silico pipeline for predicting serotypes and biotypes of Y. enterocolitica from whole-genome sequencing data using well-established molecular markers. To test the performance of this method, a dataset consisting of 242 whole genomes was constructed from the National Center for Biotechnology Information (NCBI) database; each genome had corresponding serotype information obtained from its associated metadata. The rule-based method described here uses established criteria to predict the presence of the O-antigen biosynthetic genes rfbC , wbcA , per , and wzt to identify each serotype. In addition to these markers, the presence or absence of the ail gene was used to distinguish between the O:5 and O:5,27 classifications. The accuracy and completeness of the predictions made by the pipeline were compared against those of known references. Overall, the pipeline correctly predicted the serotypes for 95.67% of the total dataset with a completeness rate of 95.45% (defined as the proportion of input genomes receiving a definite serotype call, excluding genomes classified as “unknown” or “ambiguous”). Ten discordances between metadata-derived and pipeline-predicted serotypes were observed, nine of which originated from a single BioProject. Independent 500-locus cgMLST analysis showed that all ten discordant genomes were more closely related to concordant genomes corresponding to the pipeline-predicted serotype than to those corresponding to the metadata-derived serotype, supporting probable inconsistencies in the public metadata. After excluding the nine discordant genomes from this BioProject, accuracy improved to 99.55%. Additionally, there were eight instances in which no serotype could be assigned due to the absence of identifiable rfbC , wbcA , per , or wzt genes. Biotype assignment was validated against 202 confirmed Y. enterocolitica strains, achieving 99.01% overall accuracy. These findings indicate that a rule-based, marker-gene strategy can provide a highly reliable and easily scalable alternative to traditional methodologies for predicting serotypes and biotypes, which must be interpreted jointly for accurate pathogenicity assessment from large sets of genomic sequences. Moreover, it demonstrates the feasibility of integrating in silico approaches into genomics surveillance pipelines and emphasizes the need for high-quality metadata in publicly available databases.