A procedure for the statistical evaluation of Ames Salmonella assay results. Comparison of results among 4 laboratories

Ames Salmonella test data collected in our laboratory and 3 National Cancer Institute contract laboratories were analyzed to study the distribution of experimental errors associated with the test. It is shown that the Poisson distribution is not appropriate, and that the power transformation model Y = (revertants/plate)lambda, with lambda = 0.2 as estimated by the methods of Box and Cox, produced a measurement scale on which the experimental errors could be adequately described by a normal (Gaussian) distribution with a constant variance. The modeling procedure enables one to properly use analysis of variance, regression analysis, and Student's t test to analyze Ames Salmonella test results, and well-known statistical quality control procedures to monitor laboratory performance. The method detects weak mutagenic activity and measures the amount and uncertainty of the increase in revertants/plate. The development of the power transformation model is discussed and examples of its use in the interpretation of Ames Salmonella assay results are included.