RNAlight

RNAlight predicts RNA subcellular localization and identifies nucleotide determinants using a LightGBM-based machine learning model.

Key Features:

• LightGBM classification: Uses LightGBM to classify RNAs by subcellular localization.
• Nucleotide k-mer features: Extracts nucleotide k-mers as input features representing sequence determinants.
• Tree SHAP interpretation: Applies Tree SHAP to quantify the contribution of k-mers to localization predictions.
• Assembly of k-mers: Assembles predictive k-mers into broader sequence features.
• RBP motif mapping: Maps assembled sequence features to known RNA-binding protein (RBP)-associated motifs.
• Multi-RNA class support: Predicts localization for long non-coding RNAs (lncRNAs), messenger RNAs (mRNAs), small nuclear RNAs (snRNAs), small nucleolar RNAs (snoRNAs), and circular RNAs.
• Compartmental distinction: Differentiates cytoplasmic versus nuclear localization.

Scientific Applications:

• lncRNA and mRNA localization analysis: Characterizes nucleotide determinants of cytoplasmic versus nuclear localization for lncRNAs and mRNAs.
• snRNA, snoRNA, and circular RNA prediction: Predicts subcellular localization of snRNAs, snoRNAs, and circular RNAs.
• RBP interaction studies: Links sequence features to RNA-binding proteins to investigate molecular interactions governing localization.
• Sequence motif discovery: Identifies k-mers and motifs that underlie differential RNA localization patterns.

Methodology:

Performs LightGBM classification on nucleotide k-mer features, applies Tree SHAP for feature attribution, assembles k-mers into broader sequence features, and maps these features to RBP-associated motifs.