RF-PseU

RF-PseU predicts pseudouridine sites in RNA to enable accurate identification of RNA pseudouridylation for functional analysis.

Key Features:

• Target: Predicts pseudouridine (Ψ) modification sites within RNA sequences.
• Core algorithm: Implements a Random Forest classifier as the predictive model.
• Feature optimization: Employs the light gradient boosting machine algorithm (LightGBM) together with an incremental feature selection strategy to identify the optimal feature-space vector.
• Feature representation: Refines feature representation to improve predictive performance.
• Benchmarking: Evaluated using benchmark data sets from three species.
• Performance: Reports integrated average leave-one-out cross-validation accuracy of 71.4% and independent testing accuracy of 74.7%, representing improvements of 3.63% and 4.77% over the best existing predictor.

Scientific Applications:

• Pseudouridine site identification: Locating pseudouridylation sites in RNA to support studies of RNA modification landscapes.
• Functional investigation: Supporting analysis of the roles of pseudouridylation in cellular biological and physiological processes.
• Method benchmarking: Comparing predictive performance of computational methods across species-specific benchmark data sets.

Methodology:

Uses a Random Forest model trained on features selected via the light gradient boosting machine algorithm (LightGBM) with an incremental feature selection strategy; performance evaluated by integrated average leave-one-out cross-validation and independent testing on benchmark data sets from three species.