Fig. 2
Models perform better than expected at random. Observed −log10(p value) versus the null expectation for submitted predictions for predicting a presence of symptoms (SC2) and b log symptom score (SC3), where p values were assessed by permutation of the predictions relative to the true values. For both subchallenges significant enrichment of p values (Kolmogorov–Smirnov test for enrichment p value 0.008, 0.002, 0.021, and 0.05 for AUPR(T0), AUROC(T0), AUPR(T24), and AUROC(T24), respectively, for presence of symptoms, and one-sided Kolmogorov–Smirnov test for enrichment p value 0.005 and 0.035 for T0 and T24, respectively, for log symptom score) across submissions demonstrates that pre-exposure and early post-exposure transcriptomic data can predict susceptibility to respiratory viruses
