Artificial neural networks and multivariate approaches to identifying donor parents for yield and downy mildew tolerance in cucumber ( Cucumis sativus L.)

Developing productive cucumber varieties that exhibit tolerance to downy mildew is a breeders’ priority. The study was conducted using twenty indigenous cucumber genotypes to identify donor parents for future breeding programs, having higher yield in addition to downy mildew disease tolerance. Sensitivity analysis using artificial neural network revealed that number of fruits per plant, downy mildew percent disease index, and fruit weight significantly influence fruit yield per plant. The percent disease index of downy mildew was primarily influenced by the vitamin C content of the fruit, followed by internodal length and number of primary branches per plant. Training and testing root mean square error values are low indicating an excellent model fit. High genotypic coefficient of variation, heritability, and genetic advance for key traits indicate additive gene action, suggesting effective improvement through direct selection without progeny testing. Five groups of genotypes were created using Mahalanobis’ squared distance statistics. The biplot obtained from principal component analysis showed high variability among genotypes. Based on multivariate analysis, sensitivity analysis, downy mildew severity traits and per se performance of genotype, seven genotypes were acknowledged as potential donors that could be passed on to the breeders.