IDENTIFICATION OF BACTERIA BY ARTIFICIAL NEURAL NETWORKS

To assess the potential for applying artifical neural networks to the identification of Gram-negative rods, four different networks were tra ined, differing by the number quality of strains in the training and t est sets. A network was trained using 3429 strains and tested with 685 9 strains identified 95.5% strains at the species level and 97.7% stra ins at the genus level. The sensitivity was 95.5 and specificity 99.9% . It was concluded that artificial neural networks are useful tools fo r phenotypic identification of Gram-negative rods.