Porpoise

Porpoise predicts pseudouridine sites from RNA sequence data to enable in silico identification of pseudouridine modifications for studying their roles in RNA biology.

Key Features:

• Feature Encoding Schemes: Evaluates 18 feature encoding schemes including binary features, pseudo k-tuple composition, nucleotide chemical properties, and position-specific trinucleotide propensity for single-strand sequences (PSTNPss).
• Stacked Ensemble Learning Framework: Integrates selected features into a stacked ensemble learning framework that combines multiple machine learning models to improve predictive accuracy and robustness.
• Performance Evaluation: Assesses performance via rigorous cross-validation on benchmark datasets and independent testing, with model interpretation tools highlighting the critical role of PSTNPs in prediction accuracy.

Scientific Applications:

• Prediction of pseudouridine sites: Generates candidate pseudouridine sites across diverse RNA types from sequence data.
• Support for functional studies: Provides predictions that can be used to formulate biological hypotheses and investigate the cellular roles of pseudouridine.

Methodology:

Evaluates 18 feature encoding schemes (including binary features, pseudo k-tuple composition, nucleotide chemical properties, PSTNPss), selects features for a stacked ensemble learning framework, evaluates models by cross-validation on benchmark datasets and independent testing, and applies model interpretation tools to assess feature contributions.