Downy mildew of grapevines is one of the most destructive diseases caused by an obligate biotrophic oomycete Plasmopara viticola, triggering severe yield loss. Regular applications of fungicides are necessary to prevent such losses, but this leads to severe environmental issues and decreased social acceptance. As a potential equivalent to traditional downy mildew management strategies, cultivars with durable resistance could contribute to sustainable and environmentally friendly viticulture. It is, therefore, common for grapevine breeders to develop fungus-resistant varieties utilizing naturally occurring resistance from wild species. Therefore, the primary breeding goal is to identify new resistances with different defence mechanisms and stack them in new varieties to prevent disease outbreaks and resistance-breaking isolates, thus minimizing fungicides in viticulture. Heretofore, more than 30 resistance loci against P. viticola have been already discovered. In this study, the bi-parental F1 population (‘Morio Muskat’ x COxGT2 (V. coignetiae x ’Gewürztraminer’)) Gf.2018-063 was investigated to identify and map resistance to P. viticola. The source of resistance is the East Asian wild species Vitis coignetiae. In addition to 109 simple sequence repeats (SSR) markers, 647 transferrable RNase H2-dependent amplicon sequencing (rhAmpSeq) markers are implemented in the creation of a genetic map. The resulting high-resolution rhAmpSeq map spanned a total map length of 1147.36 cM, comprising 19 linkage groups and an average distance between loci of 3.2 cM. Using each linkage map separately and three years of leaf disc assay based phenotypic data, quantitative trait locus (QTL) analysis was performed resulting in a consistent and highly significant QTL on chromosome 14 with an explained phenotypic variance of up to 36.4 %. This QTL does not share any SSR marker alleles with the pre-existing East Asian V. amurensis derived Rpv8 and Rpv12 QTLs on chr. 14. Therefore, it was designated as Rpv32 (Resistance Plasmopara viticola 32) (Malagol et al., 2023, in preparation). SSR and rhAmpSeq markers identified in this research work can be exploited in Marker-assisted selection (MAS) for introgression of Rpv32 into breeding lines and stacking resistances. Furthermore, microscopic staining studies at various time intervals and quantitative analysis of P. viticola (5 dpi) demonstrated and confirmed that the genetically identified resistant parental genotype (COxGT2), in contrary to the second parental genotype (‘Morio Muskat’), prevents pathogen proliferation. Moreover, the population utilized in this study showed segregation for the morphological trait leaf hair and a significant QTL was identified on LG 5 with an explained variance of 24 % (ribbon trichome). The hypothesis that leaf hair serves as physical barrier against P. viticola was tested in all three years. However, no strong association was observed between the leaf hair density and P. viticola infection on the leaf discs. When different people work on phenotypic data evaluation in various years, traditional phenotyping methodologies turn out not only to be time-consuming and labor-intensive, but also immensely subjective. This subjectivity tends to introduce statistical noise and bias into the final analytical result. Therefore, this research also focused on training and developing a high- throughput SCNN (shallow convolutional neural network) based model for downy mildew disease quantification (Zendler et al., 2021). The model achieved an overall prediction accuracy of 97 %. The SCNN model performance was demonstrated by a strong and significant correlation with independently evaluated experts’ data. This SCNN model in combination with an automated imaging system, shows accuracy and potential reduction in time spent on phenotyping. This pipeline serves as a valuable tool in grapevine breeding research. As an additional aspect of this research, a Residual Networks-based Convolutional Neural Network (ResNet-CNN) for leaf hair quantification was developed due to the lack of accurate and precise tools available (Malagol et al., 2023, in preparation). The model achieved an overall prediction accuracy of 95.41 %. The validation and cross validation with two expert and two non-experts showed exceptional correlation (R = 0.98 and R = 0.92, RMSE 8.20 and 14.18, respectively). The absolute errors calculated clearly indicated bias introduced due to the subjectivity. To conclude, the developed ResNet-CNN is capable of enhancing objective phenotyping accuracy for leaf hair density, allowing for a more precise analysis of this trait (refer to Annex III & Annex IV).weiterlesen