ncDLRES

ncDLRES predicts non-coding RNA (ncRNA) families by learning sequence-derived features to classify ncRNAs for functional and comparative analyses.

Key Features:

• Dynamic LSTM feature extraction: Uses Dynamic Long Short-Term Memory networks to extract features directly from ncRNA sequences by capturing long-range dependencies and dynamic patterns.
• ResNet classification: Employs Residual Neural Network architecture to classify extracted features, using residual learning to mitigate vanishing gradients.
• Sequence-based learning: Operates directly on sequence features, bypassing reliance on predicted secondary structures.
• Reduced data requirements: Eliminates the need for consensus secondary structure annotations required by homologous sequence alignment methods.
• Pseudoknot independence: Does not depend on explicit pseudoknot modeling for prediction.
• Improved predictive performance: Demonstrates enhanced performance relative to methods that rely solely on sequence or predicted structural features.

Scientific Applications:

• ncRNA family prediction: Classifies ncRNAs into families to support annotation and comparative studies.
• Analysis without secondary structure annotations: Applicable to datasets lacking consensus secondary structure information.
• Datasets with pseudoknot complexity: Applicable where pseudoknot modeling is absent or incomplete.
• Functional inference: Supports downstream inference of ncRNA functions based on family assignments.

Methodology:

Dynamic LSTM networks extract sequence-derived features by capturing long-range dependencies and dynamic patterns, and a ResNet classifies these features using residual learning to mitigate vanishing gradients while bypassing predicted secondary-structure inputs.