JEDI

JEDI predicts circular RNAs (circRNAs) from primary nucleotide sequences by modeling splice-site junctions and their interactions to identify backsplicing events.

Key Features:

• Nucleotide-sequence-only input: Operates end-to-end using only the primary nucleotide sequence as input for circRNA prediction.
• Junction encoding: Encodes each splice junction site using deep bidirectional recurrent neural networks with attention mechanisms.
• Cross-attention modeling: Implements a cross-attention layer to model interactions among donor and acceptor splice sites to facilitate backsplicing prediction.
• End-to-end framework: Provides an integrated computational framework that directly maps sequence input to circRNA predictions.
• Interpretability of splice-site relations: Interprets relationships among splice sites and identifies hotspots for backsplicing within gene regions.
• Multi-level prediction: Performs both isoform-level and gene-level circular RNA prediction.
• Zero-shot backsplicing discovery: Enables zero-shot discovery of previously undetected backsplicing events.
• Performance: Demonstrated improved performance relative to several state-of-the-art approaches in experimental evaluations.

Scientific Applications:

• Circular RNA identification: Prediction of circRNAs and backsplicing events from sequence data.
• Splice-site interaction mapping: Inferring interactions between donor and acceptor splice sites and mapping backsplicing hotspots.
• Isoform and gene annotation: Annotating circRNA isoforms and assessing gene-level circRNA presence.
• Discovery of novel splicing events: Zero-shot discovery of previously undetected backsplicing events.
• Functional studies: Facilitating studies of circRNA roles in gene regulation and associations with disease by providing accurate circRNA calls.

Methodology:

End-to-end framework that uses only nucleotide sequence input, encodes each junction site with deep bidirectional recurrent neural networks and attention mechanisms, and applies a cross-attention layer to model interactions among splice sites.