Accepted_test

Three datasets of bread wheat, rye and chickpea were used to compare the performance of several parametric and nonparametric methods in predicting phenotype from genomic data. The rye and bread wheat datasets were generated by the Russian Bread consortium. Genomic prediction (GP) models were constructed for 15 traits for rye and for 7 traits for wheat dataset including BLUP, Bayessian regression, Extreme Gradient Boosting. The combination of CNN for feature extraction and Extreme Gradient Boosting for regression was applied to model two productivity traits of chickpea, 1000 seed weight and number of seeds per plant. The non-parametric methods yielded superior results than parametric methods across cases. The best accuracy of 88% was achieved for the Length of third subflag leaf on the rye dataset using the Extreme Gradient Boosting model. Accurate prediction models can be generated by using CNN in combination for feature extraction, however further studies are necessary to provide new valuable and unbiased information for assessing their methodological potential for GP.