CDASOR

CDASOR predicts associations between circular RNAs (circRNAs) and diseases using convolutional neural networks (CNNs) and recurrent neural networks (RNNs), including bi-directional long short-term memory (Bi-LSTM) networks, to analyze sequence and disease-ontology representations for biomarker discovery.

Key Features:

• Sequence Encoding: Encodes circRNA sequences using continuous k-mers and transforms them into low-dimensional vectors.
• Feature Extraction: Uses 1D convolutional neural networks (1D CNNs) to extract local feature vectors from encoded sequences.
• Dependency Learning: Employs bi-directional long short-term memory (Bi-LSTM) networks to learn long-term dependencies in circRNA sequences.
• Ontology Representation: Serializes disease ontology into structured sentences, converts them into low-dimensional vectors, and captures term dependencies via neural networks.
• Association Pattern Recognition: Analyzes known circRNA–disease associations with neural networks to derive high-level representations for prediction.
• Implementation: Implemented in Python.

Scientific Applications:

• Prioritization for Validation: Predicts potential circRNA–disease associations to prioritize candidates for experimental validation.
• Biomarker Discovery: Supports identification of circRNAs as diagnostic or prognostic biomarkers, leveraging circRNA stability.
• Computational Screening: Provides in silico screening to reduce experimental burden in de novo association discovery.

Methodology:

CircRNA sequences are encoded into continuous k-mers and embedded as low-dimensional vectors; 1D CNNs extract local features; Bi-LSTMs capture long-term dependencies; disease ontologies are serialized into structured sentences and embedded to capture hierarchical term dependencies; neural networks learn patterns from known circRNA–disease associations to derive high-level representations used for prediction.