Taxonomic Probability Matrix for Use With Slowly Growing Mycobacteria

Abstract

A probability matrix for identification was developed from data derived from a series of cooperative studies of slowly growing mycobacteria. The matrix includes feature frequencies exhibited by 14 numerical taxonomy clusters in 34 tests. The clusters correspond to 13 defined species. The matrix is designed primarily to screen strains either for membership in 1 of the 14 taxa or for exclusion from these taxa and, thus, to determine whether the strains are in need of further characterization. The matrix was used in the analysis of 298 strains. Two related parameters were used as decision thresholds. First, the probability of the most likely taxon must be 99 times greater than that of the second most likely taxon. Second, the absolute likelihood of the strain being in the most likely taxon must be at least 0.01 times that of a “perfect” strain of the taxon. By using these thresholds and additional judgments, 83 strains were found to be appropriate for further study, with a likelihood that 53% of these strains belong to unrepresented taxa.