ShaKer

ShaKer predicts SHAPE reactivity profiles from RNA sequence using a graph-kernel-based machine learning approach to infer RNA structural properties without requiring reference structures.

Key Features:

• Graph-kernel-based machine learning: Uses a graph-kernel-based model to learn relationships between RNA sequence features and SHAPE reactivity.
• Sequence-only input: Predicts nucleotide reactivity directly from RNA sequence without needing pre-existing or manually curated reference structures.
• Training on experimental SHAPE data: Model parameters are learned from experimental SHAPE measurements provided as training data.
• Ensemble-aware inference: The graph-kernel approach facilitates exploration of the ensemble of possible RNA structures underlying reactivity signals.
• Performance evaluation: Model performance was assessed using accuracy and accessibility metrics compared to experimental SHAPE data and competing methods.

Scientific Applications:

• RNA structuredness analysis: Enables transcriptome-scale inference of nucleotide-level structural propensity from sequence-derived SHAPE predictions.
• RNA structure prediction and annotation: Supplies experiment-driven SHAPE annotations to inform secondary structure modeling and comparative analyses.
• RNA–RNA interaction studies: Provides reactivity profiles that can refine predictions of RNA–RNA interactions and accessible regions.
• Transcriptome-wide SHAPE prediction: Facilitates generation of SHAPE reactivity predictions across large sets of transcripts where experimental data or reference structures are lacking.

Methodology:

ShaKer applies a graph-kernel-based machine learning model trained on experimental SHAPE data to predict nucleotide-level reactivity from RNA sequence and was evaluated using accuracy and accessibility metrics.