PredCRG

PredCRG predicts plant circadian genes from protein sequence-derived compositional, transitional, and physico-chemical features using Support Vector Machine classifiers for proteome-scale circadian protein identification.

Key Features:

• Sequence-Based Prediction: Predicts circadian genes using compositional, transitional, and physico-chemical features of protein sequences rather than gene expression profiles.
• Kernel Utilization: Evaluated SVM kernels including linear, polynomial, radial, sigmoid, hyperbolic, Bessel, and Laplace, with the Laplace kernel achieving 62.48% accuracy in fivefold cross-validation and 62.96% accuracy on an independent dataset.
• Comparative Performance: Outperformed Random Forest, Bagging, AdaBoost, XGBoost, and LASSO for circadian gene identification.
• Proteome-Wide Identification: Applied proteome-wide to identify circadian proteins in Oryza sativa and Sorghum bicolor.
• Functional Annotation: Functionally annotated predicted circadian proteins using Gene Ontology (GO) terms.

Scientific Applications:

• Proteome-level circadian discovery: Enables identification of circadian proteins across whole proteomes in plant species such as Oryza sativa and Sorghum bicolor.
• Machine-learning benchmarking: Provides a sequence-based benchmark for comparing classifiers in circadian gene prediction against methods like Random Forest, Bagging, AdaBoost, XGBoost, and LASSO.
• Functional characterization: Supports downstream functional interpretation of predicted circadian proteins via Gene Ontology annotation.

Methodology:

Uses Support Vector Machine classifiers with kernels (linear, polynomial, radial, sigmoid, hyperbolic, Bessel, Laplace) on compositional, transitional, and physico-chemical protein sequence features, evaluated by fivefold cross-validation and independent dataset validation, with predicted proteins annotated by Gene Ontology.