Fig. 3

Correlation of pathogenicity-associated genotypes with in vitro pathogenicity-related phenotypes. a–e Distribution of scores for five in vitro phenotypes between asymptomatic carriage (red, n = 44) and infection S. epidermidis isolates (blue, n = 36): a interleukin-8 (IL-8) quantification in HaCaT keratinocytes and d in whole human blood serum, b biofilm formation, c cytotoxicity using a vesicle assay (comparing 17 asymptomatic with 31 infection isolates), and e methicillin resistance (defined as growth at a concentration of > 0.25 mg/L). The mean and s.d. error bars are shown for A-D with P-value determined using two tailed t test, n.s. indicates not significant. Two-tailed Fishers exact test was used to determine significant difference (P-value) in methicillin resistance (e). f Manhattan plot of Fisher’s exact test P-values correlating the prevalence of each GWAS-associated k-mer with high and low percentiles of four in vitro phenotype scores performed on the same S. epidermidis isolates used for GWAS. The red dotted line indicates the lower threshold for statistical significance used in the GWAS. The blue dotted line indicates a cut-off for top correlation values. Top values mapped to 61 genes. g–k Identification of predictive genotypes for pathogenicity in S. epidermidis using random forest (RF) models. g Importance of the top 1000 (of 1900) k-mer predictors from the primary GWAS; h predictor importance (left y-axis) among the top 20 phenotype correlated predictors. The red dotted line shows the classification accuracy (right y-axis) of the sub-models in which only the corresponding top predictors are included. i Predictor importance of the four laboratory phenotype-specific k-mers included in the final model. j Change in risk score for a specific k-mer profile when the colour-indicated k-mer is present (y-axis) compared to absent (x-axis). A point above the diagonal implies that the risk score is increased when the k-mer is present. k ROC curve showing the overall performance of the classifier