Deep learning empowers genomic selection of pest-resistant grapevine.

Area of Science:

• Genomics
• Plant Breeding
• Computational Biology

Background:

• Crop pests threaten global food security, with conventional insecticide use leading to resistance and ecological issues.
• Natural pest resistance varies in crops and wild relatives, offering potential for breeding resistant varieties.
• Genomic selection (GS) combined with advanced computational methods can accelerate the development of pest-resistant crops.

Purpose of the Study:

• To integrate deep learning (DL), machine learning (ML), plant phenomics, quantitative genetics, and transcriptomics for genomic selection of pest resistance in grapevine.
• To develop accurate methods for assessing pest damage using DL algorithms.
• To identify genetic loci and candidate genes associated with pest resistance in grapevine.

Main Methods:

• Deep convolutional neural networks (DCNNs) were used to classify and quantify pest damage on grape leaves.
• Genome-Wide Association Studies (GWAS) were performed on 231 grapevine accessions using genome resequencing data.
• Transcriptome data was integrated with GWAS results to pinpoint specific pest-resistance genes.
• Machine learning-based genomic selection models were developed to predict pest resistance.

Main Results:

• DCNNs achieved high accuracy in pest damage assessment (95.3% classification, 0.94 correlation).
• GWAS identified 69 quantitative trait loci (QTLs) and 139 candidate genes involved in plant defense pathways.
• Specific genes like *ACA12* and *CRK3* were identified as crucial for herbivore response.
• ML-based GS accurately predicted pest resistance (95.7% accuracy, 0.90 correlation).

Conclusions:

• Deep learning and machine learning are powerful tools for plant phenomics and genomic selection.
• This integrated approach facilitates the genomic breeding of pest-resistant grapevine varieties.
• The study provides a framework for developing sustainable pest management strategies in agriculture.