Use of canonical variates analysis in differentiation of bacteria by pyrolysis gas-liquid chromatography

Low-resolution pyrolysis gas-liquid chromatography (p.g.l.c.) can differentiate genera of aerobic food spoilage bacteria. Multivariate statistical techniques were applied but neither principal components nor furthest neighbour cluster analysis produced a consistent differentiation although both confirmed the reproducibility of p.g.l.c. When the distance between genera was redefined in terms of Mahalanobis D2- a generalized concept taking into account scatter around the mean-good differentiation was observed and could be displayed graphically by plotting the genus group means relative to the first two canonical variate axes. The coefficients of the canonical variates provide a strategy for discriminating between the genus groups. Some practical problems in the identification of unknowns using this technique are discussed.