IML-TYLCVs

IML-TYLCVs predicts Tomato yellow leaf curl virus (TYLCV) symptom severity (mild versus severe) from viral sequence data using machine learning for pathogenicity assessment.

Key Features:

• Feature extraction: Eleven different feature encodings, including hybrid features, were extracted from characterized TYLCV sequences isolated in Korea.
• Classifiers: Eight distinct classifiers were trained as part of the predictive framework.
• Cross-validation: Model evaluation used randomized 10-fold cross-validation.
• Model selection: The top 90 models were selected based on performance metrics.
• Ensemble-derived features: Predicted class labels from the selected models were combined to form reduced features.
• Final predictor: A multilayer perceptron (MLP) was trained on the reduced features to produce final severity predictions.
• Empirical validation: Blind predictions were performed on three newly isolated TYLCV groups from resistant tomatoes (2022) and results were confirmed by virus-challenging experiments using infectious clones.

Scientific Applications:

• Symptom severity classification: Classifies TYLCV sequences into mild or severe symptom categories.
• Variant assessment: Predicts pathogenic potential of newly isolated TYLCV groups, demonstrated for three groups from 2022.
• Experimental confirmation: Provides sequence-based predictions that were validated by virus-challenging experiments with infectious clones.
• Pathogenicity and breeding support: Supports assessment of TYLCV pathogenicity to inform strategies for developing resistant tomato lines.

Methodology:

Eleven feature encodings (including hybrid features) were extracted from TYLCV sequences; eight classifiers were trained and evaluated via randomized 10-fold cross-validation; the top 90 models were selected, their predicted class labels were combined into reduced features, and a multilayer perceptron was trained on those reduced features as the final prediction model.