IRESpy

IRESpy: XGBoost-based predictor of internal ribosome entry sites in mRNA untranslated regions

IRESpy predicts internal ribosome entry sites (IRES) in mRNA untranslated regions by classifying sequences as IRES or non-IRES using sequence and RNA structural features, supporting high-throughput scanning such as analysis of human 5' UTRs.

Key Features:

• XGBoost classifier: Predicts IRES versus non-IRES sequences using an XGBoost model optimized for accuracy and computational speed.
• Sequence kmer features: Encodes sequence-based kmer words as discriminative inputs for classification.
• RNA structural features: Uses structural characteristics including QMFE (Minimum Free Energy) as model inputs.
• Hybrid sequence–structure features: Combines sequence and structure information as integrated discriminators.
• Feature reduction: Reduces required features using global kmer and structural attributes to increase processing speed for high-throughput use.
• Model interpretability: Explains feature contributions using LIME (Local Interpretable Model-agnostic Explanations) and SHAP (SHapley Additive exPlanations).

Scientific Applications:

• Human 5' UTR scanning: Scans human 5' untranslated regions (UTRs) to identify novel IRES segments.
• Gene annotation: Supports genomics workflows by enabling IRES-informed annotation of transcripts.
• Differential gene expression analysis: Provides IRES predictions that can be incorporated into differential gene expression analyses.

Methodology:

Uses an XGBoost classifier trained on discriminative features separating IRES and non-IRES sequences, including sequence-based kmer words, RNA structural characteristics such as QMFE (Minimum Free Energy), and hybrid sequence–structure features; reduces the feature set using global kmer and structural attributes and reports feature-level explanations using LIME and SHAP.