PseU-ST

PseU-ST predicts pseudouridine (Ψ) sites in RNA sequences to identify modification locations for comparative and species-specific analyses.

Key Features:

• Position-aware sequence descriptors: capture positional nucleotide context associated with pseudouridine sites.
• Position-specific trinucleotide propensity based on single strand (PSTNPss): captures positional trinucleotide context associated with Ψ sites.
• Position-specific of three nucleotides (PS3): represents local nucleotide arrangement patterns that contribute to discrimination of modified versus unmodified sites.
• Six optimal RNA sequence encoding schemes (iLearnPlus): selected via systematic evaluation with the iLearnPlus software across a broad set of encoding strategies.
• Feature selection: chi-square filtering and incremental feature selection are used to optimize features for each encoding scheme.
• Machine learning base classifiers: four machine learning algorithms were selected based on preliminary benchmarking and used as base classifiers.
• Stacking ensemble and species-specific optimization: a stacking framework integrates base classifiers and selects optimal feature and classifier combinations for Homo sapiens, Saccharomyces cerevisiae, and Mus musculus through performance comparisons.

Scientific Applications:

• Comparative annotation of pseudouridylation sites: enables comparison of Ψ site distributions across organisms.
• Species-specific annotation: provides species-level prediction models for Homo sapiens, Saccharomyces cerevisiae, and Mus musculus.
• Analysis of RNA modification patterns and functional roles: supplies candidate Ψ sites for downstream analyses of modification patterns and their potential functions.
• Support for mechanistic studies and bioinformatics workflows: facilitates integration of predicted Ψ sites into mechanistic studies in RNA biology and related computational analyses.

Methodology:

Six optimal RNA sequence encoding schemes were selected via systematic evaluation with iLearnPlus; features for each encoding were optimized using chi-square filtering and incremental feature selection; four machine learning algorithms were chosen as base classifiers and integrated via a stacking framework with species-specific selection of optimal feature and classifier combinations for Homo sapiens, Saccharomyces cerevisiae, and Mus musculus through extensive performance comparisons.