Evaluation of salinity tolerance in wheat: a novel approach using artificial neural networks and rank sum-integrate selection index methods

The prediction of grain yield (GY) is one of the most important breeding objectives in agricultural research. The aim of this study was to predict GY in wheat under both non-stress and salt-stress conditions using physiological, morphological, and phonological parameters. An artificial neural network (ANN) was trained to predict GY using a multilayer perceptron model and compare the performance of ANN models with multiple linear regression (MLR) models. For these purposes, an α-lattice design was used to study 110 wheat genotypes under non-saline and saline stress conditions (EC of 2 and 10 ds m⁻¹, respectively). Our results suggest that the Iranian wheat germplasm exhibits high genetic diversity for all studied traits. The ANN model with R² values of 0.98 and 0.95 under non-stress and saline stress conditions was a more accurate tool than MLR for predicting seed yield. According to the sensitivity analysis, biological yield and harvest index were identified as the most effective traits in GY. Therefore, these traits, along with GY were used to evaluate and screen salinity-tolerant wheat genotypes through rank sum and develop an integrated selection index. Nine promising advanced lines (No. 2, 3, 5, 7, 8, 10, 11, 12, and 13) and one tolerant cultivar (No. 31) was identified as the most salinity tolerant genotype. Overall, by selecting genotypes based on the rank sum and the developed integrated selection index in a field breeding experiment, favorable wheat genotypes can be identified for non-stress and saline stress conditions.