NetUTR

NetUTR predicts splice sites within the 5' untranslated regions (5' UTRs) of human genes to identify splicing signals in non-coding sequences and improve annotation of UTR exon-intron structure.

Key Features:

• Focus on Non-Coding Regions: Trained to recognize splice sites within "pure" UTRs while avoiding partial overlaps with protein-coding regions.
• Neural Network Algorithms: Employs neural network algorithms that are analyzed and optimized specifically for predicting splice sites in 5' UTRs.
• Correlation Analysis: Investigates correlations between 5' UTR splice sites and features in adjacent exons and introns, including compositional and positional biases.
• Compositional Bias Detection: Identifies over-representation of cytosine and guanine at the ends of non-coding exons and introns.
• Enhanced Predictive Power: Isolates splicing signal from coding noise to improve prediction accuracy for 5' UTR splice sites and outperforms NetGene2 and GenScan by a factor of 2–3 in these regions.
• Versatility Across Regions: Models trained on 5' UTRs retain predictive performance when applied to protein-coding regions, indicating shared local splicing patterns.

Scientific Applications:

• Gene structure annotation: Prediction of 5' UTR splice sites to refine exon-intron boundaries in gene models.
• Regulatory mechanism analysis: Investigation of splicing patterns in non-coding UTRs to study regulatory mechanisms governing gene expression.
• Non-coding sequence function: Analysis of the role of 5' UTRs and other non-coding sequences in genetic regulation.

Methodology:

Training and evaluation of neural network algorithms on "pure" 5' UTR sequences, combined with correlation analyses of splice sites and adjacent exon/intron features and detection of compositional and positional biases (notably cytosine and guanine enrichment).