Artificial neural network based identification of environmental bacteria by gas-chromatographic and electrophoretic data

Chemotaxonomic identification techniques are powerful tools for environment al micro-organisms, for which poor diagnostic schemes are available. Whole cellular fatty acid methyl esters (FAME) content is a stable bacterial prof ile, the analysis method is rapid, cheap, simple to perform and highly auto mated. Whole-cell protein is an even more powerful tool because it yields i nformation at or below the species level. The description of new species an d genera and subsequent continuous rearrangement provide large amounts of d ata, resulting in large databases. In order to set up suitable software too ls to work on such large databases artificial neural network (ANN) based pr ograms have been used to classify and identify marine bacteria at genus and species levels, starting from the fatty acid profiles and protein profiles respectively. We analysed 50 certified strains belonging to Halomonas, Mar inomonas, Marinospirillum, Oceanospirillum and Pseudoalteromonas genera. Bo th supervised and unsupervised ANNs provide a correct classification of the marine strains analyzed. Moreover, a set of 73 marine fresh isolates were used as an example of identification using ANNs. We propose supervised and unsupervised ANNs as a reliable tool for classification of bacteria by mean s of their FAME and of whole-protein analyses and as a sound basis for a co mprehensive artificial intelligence based system for polyphasic taxonomy. ( C) 2000 Elsevier Science B.V. All rights reserved.