Table 1.
Number of experiments by serum pool and viral strain combination.
Figure 1.
Plaque reduction estimates for each experiment.
Each black dot represents the mean reduction in plaques formed for that dilution from two repeats. The red dots are the overall means across all the experiments. Superimposed are fitted models using a probit transformation, a cloglog transformation and a non-parametric spline (the estimate of the ‘true’ titers).
Figure 2.
Plaque reduction proportions for individual dilutions.
Each solid line represents the 2.5% and 97.5% quantiles of plaque reductions from the average of two repeats from a single dilution from a particular virus – serum pool combination. The shaded area represent asymptotic 95% confidence intervals calculated from the overall variance observed in the plaque reduction proportions.
Figure 3.
Estimates of (A) bias2, (B) variance and (C) mean squared error in titers from a single set of dilutions by PRNT evaluation point for the different models.
Bias for each experiment was calculated by comparing the model PRNT results with that from a smooth spline from all experiments from that particular virus and serum pool. Only assays where titers could be calculated for all four models (63% of all assays) were used.
Table 2.
Estimated standard deviation and bias in PRNT75 estimates using the different models.
Figure 4.
(A) Variance in absolute titers by the number of repeat sets of dilutions and where the repeats are performed (either the same plate or on different plates).
All titers estimated using cloglog regression. (B) Observed titer variability. Each solid line represents the 2.5% and 97.5% quantiles from the titers estimated from the serum pool – virus combinations listed in Table 1. The shaded area represents asymptotic 95% confidence intervals calculated using a single variance estimate from all assays (represented by the yellow circle in panel A).
Figure 5.
(A) Probability of detecting a greater than four-fold difference in titers between paired sera where no true difference exists. All titers calculated through cloglog regression.
(B) Example confidence intervals for the log difference between titers calculated from paired sera (and the equivalent intervals for the ratio) under different scenarios of PRNT evaluation points and where repeat dilutions are performed.
Table 3.
Results of multilevel model for impact of experimental factors on PRNT75 estimates using cloglog regression.