Statistical interpretation of the Amerithrax "morph" assay results

In the Amerithrax investigation PCR-based "morph assays" were used to link the anthrax letters with the RMR-1029 flask at USAMRIID. Quantitative data reported for several of these assays are not consistent with Poisson sampling statistics, but instead exhibit "Taylor's Law" behavior where the variance greatly exceeds the mean. A plausible statistical model for this behavior can explain the large number of observed negative and "inconclusive" findings, and implies a high likelihood that a repository sample could contain a "morph" mutant at concentrations well above the nominal detection limit but nonetheless give a negative or inconclusive test result. A Bayesian framework relates the assay results to the probability that a sample actually contains all four morph mutants, even though it tested negative for at least one. The analysis implies that the observed false negative rate actually does not significantly weaken the conclusion that the morph assays correctly excluded all but the stocks derived from RMR-1029 as possible sources of the letter powders, at least when the test results were unambiguous. These findings expand upon and resolve some of the issues cited in recent reviews, and indicate the importance of developing a rigorous statistical framework for interpreting "morph" assay data.