Stack-DHUpred

Stack-DHUpred predicts dihydrouridine (DHU) modification sites in tRNA, mRNA, and snoRNA to support studies of RNA modification and epigenetic regulation.

Key Features:

• Stacked ensemble learning: Employs a stacking approach within ensemble learning to integrate multiple baseline predictors into a final model.
• Baseline model composition: Constructs 66 baseline models corresponding to six machine learning classifiers combined with eleven feature encoding techniques.
• Single-feature models: Trains single-feature baseline models for each feature encoding and classifier combination prior to stacking.
• Model selection: Identifies the optimal combination of baseline models to assemble the final stacked predictor.
• Independent evaluation: Assesses predictive performance on an independent dataset and reports superior accuracy compared to existing predictors.

Scientific Applications:

• Genome-wide DHU annotation: Enables annotation of DHU modification sites across genome-wide RNA datasets.
• Post-transcriptional regulation studies: Facilitates investigation of the role of DHU in post-transcriptional regulatory mechanisms.
• Disease and pathogenesis research: Supports research into the implications of DHU modifications in eukaryotic health and disease.

Methodology:

Train 66 baseline models (six classifiers × eleven feature encodings) as single-feature models, combine selected baseline models into stacked ensemble models via a stacking approach, and evaluate the final stacked model on an independent dataset.