DeepCirCode

DeepCirCode predicts circular RNA (circRNA) formation via back-splicing from nucleotide sequences using convolutional neural networks (CNNs) to identify sequence motifs involved in circRNA biogenesis.

Key Features:

• Implementation: Implemented as an R package that applies deep learning to nucleotide sequence data.
• Deep Learning Model: Employs a convolutional neural network (CNN) architecture that outperforms support vector machines and random forests in predicting back-splicing events.
• Sequence Motif Identification: Learns and represents sequence motifs that are crucial for back-splicing and align with known human sequences involved in splicing, transcription, or translation.
• Cross-Species Conservation: Identifies motifs conserved across species, including humans, mice, and fruit flies.
• Tissue- and Developmental Stage-Specific Expression: Enables analysis of circRNA expression patterns that vary by tissue type and developmental stage.

Scientific Applications:

• Predicting back-splicing events: Predicts potential back-splicing sites and circRNA formation from nucleotide sequences.
• Candidate circRNA discovery: Identifies candidate circRNAs associated with biological processes and diseases.
• Regulatory mechanism inference: Uncovers sequence features (motifs) that influence circRNA formation to inform regulatory mechanisms of gene expression.
• Comparative conservation analysis: Supports analysis of motif distribution and conservation across species.
• Tissue and developmental studies: Assists investigation of tissue- and developmental stage-specific circRNA expression patterns.

Methodology:

The method trains a CNN on nucleotide sequences to predict back-splicing sites, represents learned features as sequence motifs, and analyzes motif distribution and conservation across species.