NetGene2

NetGene2 predicts intron splice sites in DNA sequences of Arabidopsis thaliana using artificial neural networks combined with a rule-based system to improve prediction accuracy.

Key Features:

• Two-Step Prediction Scheme: Uses a global assessment of coding potential followed by localized splice-site prediction with refined cutoffs based on splice site confidence values, prediction scores, and coding context.
• Rule-Based Refinement: Applies rule-based adjustments that consider distances between potential splice sites and contextual factors and leverages non-local interactions to reduce false positives.
• Error Analysis and Novel Discoveries: Neural-network error analysis revealed T-tract prolongations containing cryptic acceptor sites at the 5' end of exons.
• Comparative Superiority: Demonstrated higher performance than GeneFinder, Gene-Mark, and Grail in benchmarks and identified the donor site of jellyfish Green Fluorescent Protein (GFP) in Arabidopsis thaliana transformants at experimentally observed positions.
• Applications to Alternative Splicing: Predicts alternatively spliced genes across dicots, monocots, and algae to support studies of gene expression and regulation.

Scientific Applications:

• Gene Prediction: Identifies splice sites to aid genomic sequence annotation and functional genomics studies.
• Alternative Splicing Analysis: Predicts alternative splicing events to support research on post-transcriptional gene regulation and protein diversity.
• Comparative Genomics: Facilitates comparative studies across plant species to elucidate evolutionary relationships and functional conservation.

Methodology:

Uses artificial neural networks and a rule-based system; performs a global assessment of coding potential; conducts localized splice-site prediction with refined cutoffs; applies rule-based adjustments considering distances and contextual biological factors; models non-local interactions among predicted splice sites to minimize false positives.