Limiting dilution assays. Experimental design and statistical analysis.

Two issues in limiting dilution analysis are considered. The first concerns the experimental design: a mathematical algorithm has been developed which calculates the number of replicate culture groups, and the (mean) number of cells per well to be used on the basis of the experimenter's a priori information about the unknown frequency. The procedure guarantees useful data if the a priori interval estimate of the frequency to be determined is correct and the cells are willing to grow. The second issue concerns the statistical method to be used for estimation of the unknown frequency. Several methods (minimum chi-square, maximum likelihood and the jackknife version of the maximum likelihood method) have been evaluated with artificial data from extensive Monte Carlo experiments. All three methods were useful in the statistical analysis of data. As a result of these experiments and theoretical considerations the jackknife version of the maximum likelihood estimation procedure is proposed as the statistical procedure of choice. The next best method is the maximum likelihood procedure.