Evaluation of Triticale Lines Compared to Wheat Cultivars in Terms of Agronomic Traits Using Supervised Learning Methods and Multivariate Statistics

Sara Farokhzadeh, Hossein Shahsavand Hassani, Zahra Zinati, and Mahdiyeh Rajaei

Abstract

The first man-made cereal grain triticale merges the yield and quality potential of wheat with tolerance of rye against environmental stresses. To find the traits with the greatest impact on the superiority of triticale over wheat, four promising triticale lines and six bread wheat varieties were evaluated in a randomized complete block design (RCBD) with three replications in the field of the Agricultural Research Station of Shiraz University, Iran. Attribute weighting algorithms revealed main leaf width (MLW), grain number per spike (GN/S), spikelet number per spike (SN/S), 1000-grain weight (TGW), plant height (PHT), main leaf length (MLL), and grain yield (GYLD) as the most discriminative traits between triticale lines and wheat cultivars. According to the principal component analysis, the first two components explained 33.46% and 24.79% of the total variation in traits, respectively, while the first component was more positively correlated to GYLD, TGW, GN/S, SN/S, leaf number (LN), and MLW. Stepwise regression study showed that grain yield as a dependent variable has been modeled as a function of the independent variables GN/S, days to heading (DHE), and SN/S. Triticale and wheat clusters, respectively, were created using cluster analysis for genotypes. Integrating results of supervised learning methods and multivariate statistics indicated that GN/S and SN/S traits can be used to create a selection index for high yield. Furthermore, the M45 triticale line with the highest value of harvest index, GYLD, and GN/S can be introduced as a promising line for high-yield production. This study paved the way to specify the key grain yield-related traits being subsequently used to increase the yield of bread wheat cultivars and triticale lines.