Deep Splicing Code (DSC)

Deep Splicing Code (DSC) classifies alternative splicing events from exon-junction sequence information using deep learning to model sequence determinants of pre-mRNA splicing.

Key Features:

• Splicing event classification: Classifies alternatively skipped exons, alternative 5' splice sites (5'ss), alternative 3' splice sites (3'ss), and constitutively spliced exons.
• Sequence input: Bases classification solely on sequence information at exon junctions.
• Deep learning framework: Employs deep learning algorithms to learn sequence patterns underlying splicing decisions.
• Predictive accuracy: Enhances accuracy in distinguishing constitutive versus alternatively spliced exons by identifying local sequence characteristics.
• Motif discovery: Captures and analyzes sequence motifs associated with competitive alternative splice sites and splicing factors.
• Motif visualization: Uses motif visualization of trained models to demonstrate identified genomic features.
• Splicing code modeling: Implements a splicing-code approach via deep learning to predict alternative splicing outcomes from sequence features.

Scientific Applications:

• Modeling alternative splicing: Improves computational models of alternative splicing behavior from sequence data.
• Predicting AS events: Provides more accurate prediction of alternative splicing event classes from exon-junction sequences.
• Gene regulation studies: Facilitates investigation of sequence determinants that regulate pre-mRNA splicing.
• Protein diversity and translational research: Informs studies of protein isoform diversity and may support potential therapeutic developments related to splicing.

Methodology:

Train deep learning models on exon-junction sequence information to classify four splicing classes, and apply motif visualization of trained models to identify sequence motifs linked to competitive alternative splice sites and splicing factors.