Fig. 4

Bacterial composition is predictive of SARS-CoV-2 status in nares, forehead, stool, and inside floor samples. The prediction performance of random forest classifiers on SARS-CoV-2 status for each sample type was assessed using AUROC (A) and AUPRC (B) for nares (n = 76), forehead (n = 79), stool (n = 44), and inside floor (n = 107), in a 100-fold cross-validation approach (see methods). C EMPress plot of the 100 features most predictive of SARS-CoV-2 status in nares, forehead, stool, and inside floor samples, where a single ASV with 100% alignment to Rothia dentocariosa was identified across all sample types. D Proportion of samples containing the highly predictive Rothia dentocariosa ASV in SARS-CoV-2 positive and negative samples from the current study and from [29] (ICU 2016 pre-COVID19)