RBPLight

RBPLight predicts plant-specific RNA-binding proteins (RBPs) to identify RBPs involved in post-transcriptional gene regulation such as splicing control, mRNA transport, and decay.

Key Features:

• Plant-specific prediction: Addresses limited generalizability of prior models trained on mammalian datasets (mice, humans) and tested on Arabidopsis thaliana to improve RBP identification in plants.
• Model ensemble: Integrates five deep learning models and ten shallow learning algorithms.
• Feature engineering: Employs 20 feature sets derived from sequence-derived and evolutionary features.
• Primary algorithm: Uses a Light Gradient Boosting Machine (LGBM) component as a standout model.
• Performance metrics: Achieved repeated five-fold cross-validation AU-ROC 91.24% and AU-PRC 91.91%, and independent dataset AU-ROC 94.00% and AU-PRC 94.50%.
• Evaluation strategy: Validated using repeated five-fold cross-validation and independent dataset testing.
• Comparative advantage: Demonstrated superior performance relative to existing state-of-the-art RBP prediction models.

Scientific Applications:

• RBP discovery in plants: Identification of plant-specific RNA-binding proteins across diverse plant species.
• Post-transcriptional regulation studies: Investigation of processes such as splicing control, mRNA transport, and mRNA decay.
• Cross-species analysis: Improves RBP prediction where mammalian-trained models (mice, humans) and models tested on Arabidopsis thaliana are insufficient.
• Model benchmarking: Provides comparative evaluation against existing state-of-the-art RBP prediction methods using AU-ROC and AU-PRC metrics.

Methodology:

Integrates five deep learning models and ten shallow learning algorithms using 20 sequence-derived and evolutionary feature sets, selects a Light Gradient Boosting Machine (LGBM) component, and evaluates performance via repeated five-fold cross-validation and independent dataset testing reporting AU-ROC and AU-PRC.