Identifying Mycobacterium tuberculosis cultures by gas-liquid chromatography and a computer-aided pattern recognition model

Abstract

A procedure that uses gas-liquid chromatography and a pattern recognition computer model was developed for distinguishing cultures of Mycobacterium tuberculosis from cultures of other mycobacteria, common bacteria, and fungi. In this procedure, a sample of a culture preparation is methanolyzed and trimethylsilylated sequentially and injected into a gas chromatograph equiped with a flame ionization detector. A pattern recognition procedure computes an error score by comparing the gas-liquid chromatography peak responses of a culture to those of a standard M. tuberculosis culture. Ten M. tuberculosis cultures were used in the development of the pattern recognition model. Computed error scores of 5 or less were established for identifying an M. tuberculosis culture. The method was evaluated with two sets of test samples, non-M. tuberculosis and M. tuberculosis cultures. Sample identification was correct for all 14 M. tuberculosis cultures (M. tuberculosis or non-M. tuberculosis), 45 fungal cultures, 94 bacterial cultures, and all but 1 of 18 cultures of mycobacteria other than tuberculosis (MOTT). The false prediction represented an isolate of M. fortuitum. For M. tuberculosis, fungal, bacterial, and MOTT cultures, the ranges of error scores were 1 to 5, 16 to 33, 13 to 34, and 4 to 26, respectively. Therefore, we have demonstrated that this diagnostic model can distinguish M. tuberculosis from non-M. tuberculosis cultures with a high degree of accuracy.