Apply neural networks to model Listeria's growth.

An artificial neural network (ANN) is an information-processing technology inspired by the way biological nervous systems, such as the brain, process information and data. The key element of this process involves the structure of the information processing system. It is composed of a large number of highly interconnected processing elements-neurons-working in unison to solve specific problems.

ANNs, like people, learn by example. An ANN is configured for an application by training it with previously known data about a process. The learning process in biological systems involves making adjustments to the synaptic connections that exist between the neurons. This is true of neural networks as well. Neural networks are being applied to an increasing number of real-world problems. Their most important advantage is in solving problems often too complex for conventional technologies.

ANN software processes and correlates data using a large number of simple processing elements. The software enables computers to make intelligent decisions. They learn by reading known input-output data sets and adapt themselves to the data. The advantage of an ANN is its ability to handle complex nonlinear relationships with ease, even when the exact nature of such relationships is unknown.

Scientists at the National Institute of Agricultural Research (INRA, Laboratory of Meat Research, Theix, F-63122 Saint-Genes Champanelle, France) have used ANNs to model the growth of strains of L. monocytogenes. The growth of 66 Listeria strains (58 L. monocytogenes strains and eight L. innocua strains) was studied using the Plackett-Burman technique. Researchers used an ANN to model growth curves of the bacteria by taking into account external variables, such as temperature, pH and water activity, as well as growth variability among the 66 strains.

The experimental data were divided into two databases. One was used to teach the ANN. This database included 44 Listeria strains. The other database containing information on 22 strains of Listeria was used to validate the software. The computations generated by the ANN were satisfactory. Simulated growth curves were similar to experimental curves. Relative differences between calculated and experimental generation times were, on an average, less than 15 %. On the other hand, relative differences between calculated and experimental lag times were, on an average, equal to 50 %.

Further information. A. Lebert or I. Desnier; phone: [phone]; fax: [phone]; URL: http://www.inra.fr.