Fig. 3: Statistical assessment of pharmacological matrices to predict PrEP efficacy.

Each column corresponds to a clinical study. The top-most rows report the data-derived violin plot reporting uncertainty in the average efficacy when taken, its mean (95% confidence interval) and the mean number of infections simulated under trial conditions. The center and bottom rows report the corresponding results derived from mechanistic modeling (see Clinical Trial Simulations in Methods), testing the hypotheses indicated by the traffic light. The traffic light depicts the mechanistically modeled scenarios, from top-to-bottom: (i) If adherence was incomplete, drug concentrations in PBMC predicted efficacy and potency was estimated based on dNTP levels in CD4+ cells (red light on), vs. (ii) when efficacy was estimated based on drug concentrations in rectal tissue and potency was estimated based on dNTP levels in rectal tissue cells (red, green and blue lights on). For each hypothesis, a one-sided P-value was computed for differences in clinical trial outcomes (number of infected individuals) between the data-derived and mechanistically modeled hypotheses, under the null hypothesis (\({{{\mathcal{H}}}}_{0}\): distributions overlap) versus the alternative (\({{{\mathcal{H}}}}_{1}\): distributions do not overlap). P-values were estimated computed by running 10⁵ simulations per pair. Box plots show the median (white dot), interquartile range (box; 25th–75th percentiles), and whiskers extending to values within 1.5 x IQR, based on 1000 samples. Source data are provided as a Source Data file. PrEP: pre-exposure prophylaxis, PBMC: peripheral blood mononuclear cells, dNTP: deoxynucleoside triphosphate, IQR: interquartile range, \({{{\rm{Inf}}}}_{sim}\): number of simulated infections.