Researchers Develop AI Model to Predict Tomato Virus Virulence Before Symptoms Appear
A team at Sungkyunkwan University has created DeepTYLCV, an AI model that predicts how severe Tomato Yellow Leaf Curl Virus (TYLCV) strains will be based on their genetic sequences alone. The model achieved 100% accuracy when tested against 15 virus isolates in controlled plant infection studies.
TYLCV ranks among the most destructive viral pathogens in tomato production globally. Severe strains cause leaf curling, yellowing, stunted growth, and significant yield losses. In recent years, highly virulent variants have spread across regions and overcome genetic resistance in some tomato cultivars.
How the Model Works
DeepTYLCV analyzes viral genome sequences rather than relying on visible plant symptoms or images. This approach allows researchers to identify mild and severe strains before symptoms develop, enabling early surveillance of emerging variants.
The model combines protein language model embeddings with a hybrid architecture that pairs a Transformer encoder and multi-scale convolutional neural network. This design captures both broad sequence patterns and local motifs associated with virulence.
Experimental Validation
The research team made blind predictions for 15 TYLCV isolates, including international reference strains and Korean field samples. They validated predictions using tomato plant infection assays, symptom severity scoring, and viral accumulation analysis.
All predictions matched the experimental results. The model correctly classified every isolate's virulence level, demonstrating practical value for identifying emerging severe variants.
Building on Previous Work
In 2023, the same team published IML-TYLCV, the first genome-based TYLCV severity prediction tool. That model was trained primarily on Korean isolates, limiting its use for globally diverse virus strains. DeepTYLCV addresses this limitation with a more generalizable framework.
Potential Applications
The model could support early viral surveillance, resistance breeding programs, and rapid assessment of newly emerging TYLCV strains. Unlike field diagnosis or symptom-based image analysis, it works independent of environmental conditions affecting plant appearance.
The study appears in Plant Communications (Impact Factor 11.6). Research teams working on AI applications in plant science may find AI for Science & Research resources valuable for similar projects.
The National Research Foundation of Korea and Sungkyunkwan University funded the research.
Your membership also unlocks:
