An electronic nose unit including 14 conducting polymer sensors, was used t
o detect the volatile profiles produced by uninoculated skimmed milk media
or that inoculated with bacteria (Pseudomonas aureofaciens, P. fluorescens,
Bacillus cereus) or yeasts (Candida pseudotropicalis. Kluyveromyces lactis
) when grown for 5 h at 30 degreesC. Using discriminant function analyses (
DFA) it was possible to separate unspoiled milk and that containing spoilag
e bacteria or yeasts. The sensor array used was a useful discriminator of m
icrobial volatile profiles. Quantitative differentiation between three diff
erent concentrations of P. aureofaciens ( 10(6), 3.5 x 10(8), 8 x 10(8) CFU
s ml(-1)) was also investigated and showed that the system could effectivel
y differentiate between treatments. Using an initial inoculum of about 10(3
)-10(4) CFUs ml(-1) it was possible to discriminate between unspoiled milk,
yeasts and bacterial species (S. aureus, B, cereus and the Pseudomonas spp
.) using principal component analyses (PCA), and also between the bacteria,
the unspoiled milk, and the two yeasts C. pseudotropicalis and K. lactis w
ith 85% of the data accounted for. The potential for differentiation betwee
n four of the five individual bacterial and yeast species was analysed afte
r 5 h growth at 25 degreesC by using a three-layer back propagation neural
network (NN) of 46 input sensor parameters. This showed that it was possibl
e to recognise, and differentiate, between species, the butanol and milk me
dium controls. Cross validation using labelled individual replicates of tre
atments as unknowns demonstrated that it was possible to differentiate betw
een (a) butanol controls; (b) unspoiled milk medium; (c) S. aureus; (d) K.
lactis; (e) C. psuedotropicalis; and (f) B. cereus. The potential for using
an electronic nose system for early detection of microbial spoilage of mil
k-based products is discussed. (C) 2001 Elsevier Science B.V. All rights re
served.