M. Giacomini et al., ARTIFICIAL NEURAL-NETWORK IDENTIFICATION OF HETEROTROPHIC MARINE-BACTERIA BASED ON THEIR FATTY-ACID COMPOSITION, IEEE transactions on biomedical engineering, 44(12), 1997, pp. 1185-1191
The traditional approach to biochemical identification of marine fresh
isolates requires considerably long culture preparation times and lar
ge quantities of expensive materials and reagents, and the results are
not very reliable. On the other hand, taxonomy tests based on DNA com
position, although sensitive and reliable, require long execution time
s and high costs. A method is presented for the classification of fatt
y-acid profiles, extracted from marine bacteria strains, at genus leve
l based on supervised artificial neural networks. The proposed method
allows the correct identification of all patterns belonging to the tra
ining set and almost all patterns belonging to the test set. Moreover,
a quantitative measure of the importance of each fatty acid for bacte
rial classification is also achieved. This measure allows the determin
ation of a cluster of fatty acids to be controlled with greater care.
The results show that the proposed method is reproducible and rapid, s
o that it can be routinely used in the marine microbiology laboratory
to identify fresh isolates.