Applying an artificial neural network approach for drought tolerance screening among Iranian wheat landraces and cultivars grown under well-watered and rain-fed conditions

Abstract

In the current study, an α-lattice design was used to investigate 320 Iranian bread wheat cultivars and landraces under non-stressed and rain-fed conditions, according to phenological, morphological and physiological parameters. An artificial neural network (ANN) was trained to evaluate the relative importance of different drought tolerance indices (DTIs) using a multilayer perceptron model. Our findings suggest that the Iranian wheat germplasm harbors large genetic diversity for all the studied traits. Correlation analyses highlighted the important role of seed number per spike, thousand kernel weight, leaf greenness and canopy temperature in predicting grain yield under both non-stressed and rain-fed conditions. Moreover, correlations between stressed-yield (Ys) and yield index (YI, r = 1**), harmonic mean (HM, r = 0.94**), geometric mean productivity (GMP, r = 0.86**), and stress tolerance index (STI, r = 0.86**) were all large, which was further confirmed by the results of ANN and a principal component analysis. A hierarchical clustering, visualized using a heatmap plot, classified cultivars and landraces into four separate groups, where high-yielding and drought-tolerant genotypes clustered in the same group. The result of ANN indicated that MP and YI had the highest relative importance for screening compatible genotypes for well-watered and rain-fed conditions, respectively. Overall, the selection of genotypes according to agronomic and physiological traits in association with an appropriate DTI can identify favorable wheat genotypes in a field trial to breed for well-watered and water-limited environments. Furthermore, the ANN successfully evaluated the relative importance of different DTIs in wheat.