ASRpro

ASRpro predicts plant abiotic stress-responsive genes and proteins using machine-learning models to identify genes associated with cold, drought, heat, light, oxidative, and salt stress.

Key Features:

• Machine Learning Algorithms: Employs Support Vector Machine (SVM), Random Forest, Adaptive Boosting (ADB), and Extreme Gradient Boosting (XGB), with SVM demonstrating superior prediction performance.
• Compositional Feature Analysis: Uses autocross covariance (ACC) and K-mer compositional features as model input to capture sequence-based patterns associated with stress response.
• Performance and Validation: Fivefold cross-validation with SVM produced accuracies of approximately 60–77% for ACC, 75–86% for K-mer, and 61–78% for combined ACC+K-mer, with consistent results on an independent dataset.
• Prediction Targets: Predicts abiotic stress-responsive genes (SRGs) and proteins across multiple stress types.
• Stress Coverage: Models were trained on datasets representing six abiotic stresses: cold, drought, heat, light, oxidative, and salt.

Scientific Applications:

• Plant Breeding: Supports identification of multistress-responsive genes to inform crop-breeding strategies for tolerance to cold, drought, heat, light, oxidative, and salt stress.
• Complementary Analysis: Complements traditional genetic approaches and transcriptome profiling by providing computational predictions that can prioritize candidate SRGs despite the labor-intensive and species-specific limitations of experimental methods.

Methodology:

Models were trained on datasets representing six abiotic stresses using autocross covariance (ACC) and K-mer compositional features as inputs; SVM, Random Forest, ADB, and XGB algorithms were evaluated using fivefold cross-validation and independent dataset testing.