Modelling molecular and inorganic data of Amanita ponderosa mushrooms using artificial neural networks

• Autor(a) principal: Salvador, Cátia
• Data de Publicação: 2011
• Outros Autores: Martins, M. Rosário, Vicente, Henrique, Neves, José, Arteiro, José, Caldeira, A. Teresa
• Tipo de documento: Artigo
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10174/3292
• topic: Amanita ponderosa; Mycorrhizal Edible Mushrooms; M13-PCR; Artificial Neural Networks

Amanita ponderosa are wild mushroom eatable, growing spontaneously in some Mediterranean microclimates, namely in Alentejo and Andaluzia, in the Iberian Peninsula, due to its Mediterranean characteristics. The aim of this study was to evaluate inorganic composition of mycorrhizal Amanita ponderosa collected from different regions of the southwest of the Iberian Peninsula and to access molecular biomarkers using artificial neural networks. Fruiting bodies of the A. ponderosa mushrooms were collected in Spring from different locations area, in the southwest of the Iberian Peninsula. Three individuals were sampled per location. The inorganic analyses showed that mineral composition of these mushrooms depends on the ecosystem where they grow. Levels of trace metals are considerably lower, acceptable to human consumption at nutritional and low toxic levels. Molecular approach using the microsatellite primer M13-PCR allowed to distinguish the mushrooms at specie level and to differentiate the A. ponderosa strains according to their location. Data mining tools were used in order to correlate inorganic and molecular results. In order to obtain the best prediction of the M13 PCR DNA band profile, different network structures and architectures were elaborated and evaluated. In the present work the error metric used was the mean squared error. The neural network selected for modelling the data has a 6-7-14 topology, i.e. an input layer with six nodes, a hidden layer with seven nodes and a fourteen nodes output layer. A good match between the experimental and predicted values can be observed.