Predicting the presence of infectious virus from PCR data: A meta-analysis of SARS-CoV-2 in non-human primates
Fig 5
The best culture model captures key sources of underlying variation in culture outcomes.
(A) All available culture data plotted against totRNA results (with vertical jitter), with all totRNA-negative samples plotted in the grey region (with horizontal and vertical jitter). (B) Distribution of median model-predicted chances of positive culture for all totRNA-positive samples, stratified by model type and observed outcomes. Samples right of the dashed vertical line are correct predictions. (C) Median predicted chance of positive culture for the simple model (top row) and all cofactor groups included in the best model (other rows) for totRNA copies (evaluated at integer values, starting at 0). Predictions were generated using the following ‘standardized cofactor set’ (which are highlighted in bold text): adult rhesus macaques inoculated with 5.5 log10 pfu and sampled at least two days post infection from inoculated tissues, where PCR targets the Nucleocapsid gene and culture uses plaque assays with VeroE6 cells. Grey boxes enclose regions where classifications differ within the cofactor group for the standardized cofactor set, as described for Fig 3C. For the simple model, it encloses regions where classifications differ between the simple model and any possible combination of cofactors. (D) 300 posterior draws from the best model for the standardized cofactor set, with colored lines as indicated in panel-specific legends. The dark blue line presents the simple model’s mean fit. Acronyms are as described in Fig 3, plus the following: E6, VeroE6; E6-SS2, VeroE6-TMPRSS2; and 76, Vero76 cells.