Segmented model as prior information for the application of artificial neural networks to classify soybean genotypes in terms of phenotypic adaptability and stability

• Ano de defesa: 2023
• Autor(a) principal: Hashimoto, Thais do Prado
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Universidade Federal de Viçosa; Genética e Melhoramento
• Palavras-chave em Português: Soja - Melhoramento genético; Mapeamento cromossômico - Métodos estatísticos; Redes neurais - (Computação); Melhoramento Vegetal
• Link de acesso: https://locus.ufv.br//handle/123456789/31644; https://doi.org/10.47328/ufvbbt.2023.577

Unlike models based on simple linear regressions, segmented models can better as- sess the adaptability and stability of genotypes, which can demonstrate a non-lin- ear pattern of response to environmental variation. Therefore, this work aimed to transpose the concepts of adaptability and stability from the statistical analysis of a segmented model to the strong discriminatory potential of an artificial neural network (ANN) and use it to classify soybean genotypes Glycine max. A total of 9,000 simu- lated soybean genotypes were previously arranged into 18 different classes, which represented the combination of nine adaptability classes by the method of Verma and collaborators (VCM) and two stability classes by the method of Finlay & Wilkinson. There was 90% agreement between the ANN and VCM analyses regarding adapta- bility classification and 20% regarding stability. With the methods presented in this work, it was demonstrated that the potential of using ANNs to evaluate the adaptability of genotypes is strong. These auxiliary parameters were used in an algorithm pro- grammed in the R software using the nnet function of the nnet package to find an ANN configuration whose maximum classification error in the testing phase was 1%. After choosing the ANN model with the smallest error, the set of real soybean genotypes was submitted to it for classification in terms of adaptability and stability. The R codes used in this manuscript are available at https://github.com/licaeufv. An ANN based on a segmented model as the VCM model were powerful to classify soybean genotypes regarding their adaptability and, possibly, can help breeders interpret data from the behavior of any cultivar in face of environmental variations considering adapted ANN models for each situation. In addition, since the stability was introduced in the ANN as a different concept from that used to classify the genotypes by the (VCM) statistical method, such classification needs to be reviewed and further improved. Keywords: Glycine Max. Artificial Intelligence. Genotypes × Environments Interaction. Data Simulation. Bioinformatics. Artificial Neural Network.