[Screening of pathogenic molecular markers of in children based on whole genome sequencing technology].

Objectives: To explore the molecular characteristics of () in children, and to compare the molecular characteristics of different types of strains (infection and colonization strains) so as to reveal pathogenic molecular markers of .

Methods: A cross-sectional study design was used to conduct nasopharyngeal swab sampling from healthy children in the community and clinical samples from infected children in the hospital. Whole genome sequencing was used to detect antibiotic resistance genes and virulence genes. A random forest method to used to screen pathogenic markers.

Results: A total of 512 strains were detected, including 272 infection strains and 240 colonization strains. For virulence genes, the carrying rates of enterotoxin genes ( and ), extracellular enzyme coding genes (, , and ), leukocytotoxin genes (, , and ) and epidermal exfoliating genes ( and ) in infection strains were higher than those in colonization strains. But the carrying rates of enterotoxin genes (, , , , , , , , and ) were lower in infection strains than in colonization strains (<0.05). For antibiotic resistance genes, the carrying rates of , , , and were significantly higher in infection strains than in colonization strains (<0.05). The accuracy of cross-validation of the random forest model for screening pathogenic markers of before and after screening was 69% and 68%, respectively, and the area under the curve was 0.75 and 0.70, respectively. The random forest model finally screened out 16 pathogenic markers (, , , , , , , , , , , , , , and ). The top five variables in the variable importance ranking were (=0.40), (=3.95), (=1.68), (=3.97), and (=1.68).

Conclusions: The random forest model can screen out pathogenic markers of and exhibits a superior predictive performance, providing genetic evidence for tracing highly pathogenic and conducting precise targeted interventions.