D-ORB

D-ORB identifies overrepresented RNA secondary-structure motifs and characterizes family-specific non-pseudoknotted secondary structures to infer the structural composition of RNA families.

Key Features:

• Alignment-free motif discovery: Detects common structural elements in functionally related RNA sequences without relying on sequence alignment.
• Secondary conformational landscape analysis: Identifies overrepresented motifs within the secondary conformational landscapes of RNA families.
• Comparative motif significance: Compares discovered motifs against those found in unrelated sequences to identify family-specific structural features.
• Non-pseudoknotted structure generation: Produces non-pseudoknotted secondary-structure models based on identified motifs.
• Machine learning integration: Applies a deep neural network classifier and two decision trees to fit and evaluate overrepresented motifs.
• Statistical inference: Uses a statistical approach to derive the structural composition of RNA families with reported high precision.
• Rfam modeling: Has been applied to model structures for more than a hundred Rfam families.
• Contrast to covariance models (CM): Addresses limitations of alignment-based covariance models (CM) that may miss motifs due to alternative folding dynamics.

Scientific Applications:

• Conserved motif identification: Detects structural motifs that are conserved among functionally related RNAs for downstream functional inference.
• Rfam family characterization: Infers non-pseudoknotted secondary structures and structural composition for Rfam families.
• Comparative structural analysis: Determines which structural features are specific to a family by comparison with unrelated sequences.
• RNA structure–function studies: Provides statistical assessments of family-level structural elements to support studies of RNA structure and function.

Methodology:

Detects overrepresented motifs within secondary conformational landscapes, compares motifs to those from unrelated sequences, generates non-pseudoknotted secondary structures from identified motifs, and applies a deep neural network classifier plus two decision trees within a statistical framework.