Grapevine Rootstock and Scion Genotypes’ Symbiosis with Soil Microbiome: A Machine Learning Revelation for Climate-Resilient Viticulture

Abstract

AbstractGiven the impact of climate change on agriculture, the development of resilient crop cultivars is imperative. A healthy plant microbiota is key to plant productivity, influencing nutrient absorption, disease resistance, and overall vigor. The plant genetic factors controlling the assembly of microbial communities are still unknown. Here we examine if Machine Learning can predict grapevine rootstock and scion genotypes based on soil microbiota, despite environmental variability. The study utilized soil microbial bacteriome datasets from 281 vineyards across 13 countries and five continents, featuring 34 differentVitis viniferacultivars grafted onto, often ambiguous, rootstocks. Random Forests, Adaptive Boost, Gradient Boost, Support Vector Machines, Gaussian and Bernoulli Naïve Bayes, k-Nearest Neighbor, and Neural Networks algorithms were employed to predict continent, country, scion, and rootstock cultivar, under two filtering criteria: retaining sparse classes, ensuring class diversity, and excluding sparse classes assessing model robustness against overfitting. Both criteria showed remarkable F1-weighted scores (>0.8) for all classes, for most algorithms. Moreover, successful rootstock and scion genotype prediction from soil microbiomes confirms that genotypes of both plant parts shape the microbiome. These insights pave the way for identifying plant genes for use with breeding programs that enhance plant productivity and sustainability by improving the plant-microbiota relationship.