Fig. 2. Clinimetric properties of species-level prediction with raspir (red) and without raspir (black).

A Average test sensitivities of 81.0% and 99.0% were observed for simulation runs with raspir and without raspir, respectively. The test sensitivity was significantly higher without raspir (Mann–Whitney–Wilcoxon, p value < 0.0001, effect size r = 0.43, confidence intervals = 0.28–0.57). However, the sensitivity was similar for all simulation runs with at least 100 reads per rare species (Mann–Whitney–Wilcoxon, p value = 1). B Average test specificities of 99.2% and 0.8% were observed for simulation runs with raspir and without raspir, respectively. The specificity with raspir was hence significantly higher (Mann–Whitney–Wilcoxon, p value < 0.0001, effect size r = 0.87, confidence intervals = 0.86–0.87). C Average false discovery rates of 1.3% and 56.7% were observed for simulation runs with raspir and without raspir, respectively. The false discovery rate of raspir was significantly lower (Mann–Whitney–Wilcoxon, p value < 0.0001, effect size r = 0.87, confidence intervals = 0.87–0.88). D Average false omission rates of 12.9% and 50% were observed for simulation runs with raspir and without, respectively. The false omission rate was significantly lower with raspir (Mann–Whitney–Wilcoxon, p value < 0.0001, effect size r = 0.92, confidence intervals = 0.91–0.94). The points and lines represent median values, the error bars show the minimum and maximum values obtained during all simulations. The individual data points can be obtained from Supplementary Table 2.