KnotShape

KnotShape predicts abstract shapes and secondary structures of non-coding RNAs (ncRNAs), improving prediction accuracy for sequences containing pseudoknots.

Key Features:

• Abstract shape representation: Preserves adjacency and nesting of structural features while ignoring helix and loop length details.
• SVM-based feature selection: Uses Support Vector Machine (SVM)-based feature selection to identify sequence and structural attributes and derive consensus abstract shapes from homologous ncRNAs.
• Consensus-guided structure prediction: Integrates derived consensus abstract shapes into structure prediction to guide secondary-structure inference, including pseudoknotted configurations.
• Benchmark performance: Evaluated on hundreds of ncRNA datasets with and without pseudoknots, reporting an 18% higher accuracy in shape prediction, ~29% increased sensitivity, and ~10% improved positive predictive value versus other tools.

Scientific Applications:

• ncRNA secondary-structure prediction: Improves prediction of native secondary structures for non-coding RNAs, including sequences containing pseudoknots.
• Consensus shape inference from homologs: Derives consensus abstract shapes from homologous ncRNA sequences for comparative structural analysis.
• Benchmarking pseudoknot prediction methods: Provides quantitative performance measures (accuracy, sensitivity, PPV) for evaluating pseudoknot-aware prediction approaches.

Methodology:

Represents structures as abstract shapes that preserve adjacency and nesting while ignoring helix and loop lengths; applies Support Vector Machine (SVM)-based feature selection to identify discriminative sequence and structural features and derive consensus shapes from homologous ncRNAs; integrates predicted consensus shapes into structure prediction and evaluates performance on hundreds of ncRNA datasets with and without pseudoknots.