Calculation of cut-off, sensitivity, specificity and accuracy

For the samples of the panel, we used receiver operating characteristics (ROC) curve analysis to determine the cut-off values for each antigen, its sensitivity, specificity and accuracy which corresponds to the area under the curve (AUC). The ROC curve analysis was performed with the Life module of XLSTAT (Addinsoft, Paris, France) implemented in Microsoft Excel. We also determined the sensitivity, specificity and accuracy by calculating the mean MFI of negative controls for each antigen, the standard deviation to the mean (SD) and using as cut-off the mean plus three times the SD because for ONNV and USUV, no or only limited positive control samples were available. We used the Wilson method [26] to calculate online the 95% confidence intervals (CI) around the proportions (http://ww3.ac-poitiers.fr/math/prof/resso/cali/ic_phrek.html).

In the absence of positive controls for NHPs samples, we analyzed the data obtained from plasma and DBS samples with different statistical methods to determine MFI cut-off values for each antigen as reported in our previous studies on ebolavirus in NHPs [27]. We used a change-point analysis with the R package “changepoint” and calculated one single shift in the arithmetic mean with the AMOC (at most one change) method. We also fitted univariate distributions to our data and defined the cut-off based on a 0.05 risk of error [28]. The set of candidate distributions was reduced with a bootstrapped skewness-kurtosis analysis [29]. Maximum likelihood estimation was performed to select the best-fit distribution based on AIC (Akaike information criterion) using the R library “fitdistrplus” [30]. The best-fit distributions were negative binomial and negative exponential distributions and both were considered in data analyses. Data were bootstrapped 10,000 times and averaged for each antigen. Analyses were done with R software version 3.3.6. We then compared the cut-off values identified by the 3 different methods and calculated their mean as a consensus cut-off that we used in this study (S4 Table). We calculated separately cut-off values for samples collected as DBS (samples from the DRC) and those collected as blood in EDTA tubes (samples from Cameroon) because of the wide disparity of blood quantity collected with DBS. We considered a sample antigen reactive if MFI was above the cut-off value. Likewise, for samples collected as feces, we calculate MFI cut-off values separately on the data generate from the dialysates (S5 Table). We considered samples positive for a given antigen if they presented MFI above the cut-off value for this antigen.