DeepCIP

DeepCIP predicts internal ribosome entry sites (IRESs) in circular RNAs (circRNAs) to identify elements that enable cap-independent translation and assess circRNA coding potential.

Key Features:

• Multimodal deep learning: Employs a multimodal deep learning approach combining multiple data modalities for IRES prediction in circRNAs.
• Sequence and structure integration: Integrates nucleotide sequence information and RNA structural information to improve predictive accuracy.
• Enhanced feature extraction: Captures intricate patterns within RNA sequences and structures that extend beyond traditional machine learning feature sets.
• Interpretability mechanism: Provides model interpretability that reveals learned sequence patterns and motifs related to circRNA translation.
• Benchmark performance: Demonstrates superior performance compared with existing comparative methods on test sets and real circRNA IRES datasets.

Scientific Applications:

• IRES identification: Predicts IRES elements within circRNAs to map potential translation initiation sites.
• Coding potential assessment: Assesses the coding capabilities of circRNAs by identifying cap-independent translation signals.
• Functional characterization of circRNAs: Aids studies that elucidate functional roles of circRNAs through detection of translational regulatory elements.
• CircRNA-based therapeutic design: Supports design efforts for circRNA-based therapeutics by identifying IRES elements relevant to translation.

Methodology:

Uses a multimodal deep learning model that integrates sequence and RNA structural information and includes an interpretability mechanism to reveal learned sequence motifs.