High genetic differentiation of grapevine rootstock varieties determined by molecular markers and artificial neural networks

Abstract

The genetic differentiation of grapevine rootstock varieties was inferred by the Artificial Neural Network approach based on the Self-Organizing Map algorithm. A combination of RAPD and SSR molecular markers, yielding polymorphic informative loci, was used to determine the genetic characterization among the rootstock varieties 420-A, Schwarzmann, IAC-766 Campinas, Traviú, Kober 5BB, and IAC-572 Jales. A neural network algorithm, based on allelic frequency, showed that the individual grapevine rootstocks (n = 64) were grouped into three genetically differentiated clusters. Cluster 1 included only the Kober 5BB rootstock, Cluster 2 included rootstocks of the varieties Traviú and IAC-572, and Cluster 3 included 420-A, Schwarzmann and IAC-766 plants. Evidence from the current study indicates that, despite the morphological similarities of the 420-A and Kober 5BB varieties, which share the same genetic origin, two new varieties were generated that are genetically divergent and show differences in performance.

Spectral-Based Classification of Genetically Differentiated Groups in Spring Wheat Grown under Contrasting Environments

“…In fact, PLS-DA tends to make a decision criterion based on easily classifiable data, which leads to an inadequate classification for outliers [ 34 ]. The genetic structure of different plant populations has been inferred by non-supervised artificial neural network methods using SNP molecular markers [ 35 , 36 ]. Unlike previous studies, our approach used supervised classification neural network-based models, which consider foliar spectral reflectance information as “attributes” and the genetic structure (previously defined with SNPs) as the “class labels”.…”

A Deep Learning Approach to Population Structure Inference in Inbred Lines of Maize

“…Artificial neural network models have been used before in order to genetically evaluate crop germplasm collections, such as maize (Ferreira et al, 2018;Kulka et al, 2018) and grapevine (Costa et al, 2020), in which the clustering analysis was based upon competitive learning-based neural networks. This alternative method was able to analyze population structure based on not only bi-allelic but also multi-allelic data (Peña-Malavera et al, 2014;Ferreira et al, 2018) and has been demonstrated to be computationally faster than MCMC methods (Nikolic et al, 2009) and does not consider the assumption of Hardy-Weinberg equilibrium in the population being studied (Ferreira et al, 2018).…”

“…InStruct is based on the Markov chain method for parameter estimation, which is computationally time-consuming with respect to other unsupervised clustering methods (Gao et al, 2007;Stift et al, 2019). It should be noted that the artificial neural networks have the advantage of being a non-parametric method, which does not require detailed information about the physical processes being modeled and is able to analyze data containing missing data (Azevedo et al, 2015;Costa et al, 2020). Interestingly, our results confirm that DeepAE neural networks provide precise results in the identification of genetically differentiated groups and the assignment of lines into subpopulations ( Table 2).…”