Machine learning reveals potato powdery scab resistance genes

Abstract

Powdery scab, caused by the soil-borne pathogen Spongospora subterranea f. sp. subterranea, is a destructive potato disease causing substantial yield losses and reduced marketability. However, tuber-specific resistance mechanisms remain poorly understood, largely due to phenotyping bottlenecks and the dynamic infection process. Here, we developed an integrated AI-omics pipeline combining deep-learning-based image analysis with machine-learning-driven transcriptomic modelling to dissect the genetic architecture of resistance. Automated disease quantification across 12 cultivars over a 118-day growth cycle revealed tuber initiation and expansion as the two critical infection windows. Integration of high-resolution phenotypic data with transcriptomic profiling and a consensus machine-learning framework prioritized 292 high-confidence candidate genes. Functional categorisation identified lignin biosynthesis as a core defence hub, with seven lignin-associated genes, including the rate-limiting enzyme CCR and three MYB transcription factors, consistently selected across complementary algorithms. Cross-dataset validation confirmed the directional robustness of the resistance signature (68.8% concordance). Virus-induced gene silencing of St4CL1 and StCCoAOMT1 compromised resistance and reduced lignin accumulation, functionally validating the pathway’s contribution to tuber defence. Beyond lignin, our pipeline ranked several poorly characterised genes, including EDNR2GH7 and DIV3B, as high-priority candidates, pointing to additional novel resistance components. This integrative framework enables objective, high-throughput resistance screening, eliminates observer bias, and provides a broadly applicable blueprint for accelerating molecular breeding in other crop–pathogen systems constrained by conventional phenotyping.