Agricultural pest management has traditionally relied on chemical insecticides, but their overuse has led to environmental contamination, health risks, and rapidly evolving pesticide resistance. Meanwhile, natural variation in pest resistance exists within crops and their wild relatives, offering valuable resources for breeding. However, resistance traits are difficult to measure accurately, as they are often scored visually using coarse categories that fail to capture continuous variation. This limits the effectiveness of genome-wide association studies and genomic selection. Advances in deep learning provide new opportunities to extract detailed phenotypic information directly from images, overcoming subjectivity and labor constraints. Based on these challenges, there is a pressing need to conduct in-depth research on AI-enabled phenotyping and genomic breeding for pest resistance.
Researchers from the Chinese Academy of Agricultural Sciences and collaborating institutions report (DOI: 10.1093/hr/uhaf128) on 7 May 2025 in Horticulture Research that deep learning can substantially improve genomic selection for pest-resistant grapevine. The team developed convolutional neural networks to automatically assess insect damage on grape leaves and combined these data with genome resequencing, genome-wide association studies, and transcriptomic analyses. By linking AI-derived phenotypes with genetic markers, the study identifies key resistance genes and demonstrates highly accurate machine-learning-based prediction of pest resistance, offering a new framework for precision breeding.
The study analyzed 231 grapevine accessions subjected to natural infestations of the tobacco cutworm, a major leaf-feeding pest. Deep convolutional neural networks were trained to classify pest damage as mild or severe, achieving over 95% accuracy, while a custom regression model generated continuous damage scores strongly correlated with human assessments. These AI-derived phenotypes enabled more precise genetic analyses than traditional categorical scoring. Genome-wide association studies identified 69 quantitative trait loci and 139 candidate genes linked to pest resistance, many involved in jasmonic acid, salicylic acid, ethylene, and calcium-mediated signaling pathways. By integrating transcriptomic data, the researchers pinpointed key defense genes, including calcium-transporting ATPase ACA12 and the protein kinase CRK3, both strongly induced during herbivore attack. Machine-learning-based genomic selection models further demonstrated high predictive power, reaching 95.7% accuracy for binary traits and strong correlations for continuous traits. Together, these results show that combining deep learning phenotyping with genomics reveals subtle resistance mechanisms and enables reliable prediction of complex, polygenic pest-resistance traits.
“This work highlights how artificial intelligence can fundamentally change plant breeding,” said the study’s senior authors. “By replacing subjective visual scoring with fast, objective deep-learning-based phenotyping, we can capture continuous variation in pest damage that was previously overlooked. When these high-quality phenotypes are integrated with genomics and transcriptomics, they reveal the true polygenic architecture of pest resistance. This approach not only improves prediction accuracy, but also allows breeders to make informed selections much earlier in the breeding cycle.”
The findings have broad implications for sustainable agriculture and crop improvement. AI-driven phenomics enables rapid, large-scale assessment of pest resistance without increasing labor costs, making it suitable for breeding programs worldwide. By identifying resistance genes and accurately predicting pest tolerance, breeders can reduce reliance on chemical pesticides while improving crop resilience. The framework established in grapevine can be readily adapted to other crops and stress traits, supporting the development of automated, data-driven breeding platforms. Ultimately, integrating deep learning, genomics, and machine learning could accelerate the creation of pest-resistant varieties essential for food security under increasing environmental pressure.
###
References
DOI
10.1093/hr/uhaf128
Original Source URL
https://doi.org/10.1093/hr/uhaf128
Funding information
This work was supported by the National Key Research and Development Program of China (No. 2023YFD2200702), the project of National Key Laboratory for Tropical Crop Breeding (No. NKLTCB202325), the National Natural Science Foundation of China (No. 32372662), and the Science Fund Program for Distinguished Young Scholars of the National Natural Science Foundation of China (Overseas) to Yongfeng Zhou.
About Horticulture Research
Horticulture Research is an open access journal of Nanjing Agricultural University and ranked number one in the Horticulture category of the Journal Citation Reports ™ from Clarivate, 2023. The journal is committed to publishing original research articles, reviews, perspectives, comments, correspondence articles and letters to the editor related to all major horticultural plants and disciplines, including biotechnology, breeding, cellular and molecular biology, evolution, genetics, inter-species interactions, physiology, and the origination and domestication of crops.