Sweet pepper yield modeling via deep learning and selection of superior genotypes using GBLUP and MGIDI

Intelligent knowledge about Capsicum annuum L. germplasm could lead to effective management of germplasm. Here, 29 accessions of sweet pepper were investigated in two separate randomized complete block design with three replications in the field condition. Fruit yield accompanied by 13 agro-morphological traits were recorded in two experiments. Genomic fingerprinting of accessions was done by using 10 ISSR primers. The convolutional neural network (CNN) models via outputs of both correlation coefficients and stepwise regression showed the high accuracy of CNN model through correlation coefficients (R² = 0.879) in predicting fruit yield of sweet pepper. Fruit thickness and fruit width were identified simultaneously as significant components in both models. Genomic best linear unbiased prediction through 65 amplified ISSR loci showed positive and high value of additive gene effect as breeding value for traits identified in the deep learning models. Among studied germplasm, G12, G13, G14, and G25 with positive and high value of breeding value especially for traits constructed the CNN models, recognized as superior genotypes. Regardless of breeding value, multi-trait genotype-ideotype distance index by utilizing all recorded agro-morphological traits simultaneously, revealed G11, G12, G13, and G15 as promising genotypes. So, G12 and G13 which have ideal values of studied traits simultaneously and also positive breeding value could be considered as promising parents for future breeding programs. The study concludes that a CNN model focusing on morphological traits with additive genetic control, combined with the MTSI index, can effectively enhance parental selection in sweet pepper.