GeneParser

GeneParser predicts protein-coding gene structures in genomic DNA by distinguishing introns and exons.

Key Features:

• Content Statistics Analysis: Evaluates subintervals in a given sequence using content statistics indicative of intronic or exonic regions.
• Neural Network Weighting: Employs a neural network to weight gathered information and approximate the log-likelihood that each subinterval represents an intron or exon (first, internal, or last).
• Dynamic Programming Optimization: Uses a dynamic programming algorithm to identify the combination of introns and exons that maximizes the likelihood function derived from the weighted data.
• Ranked Suboptimal Solutions: Rapidly generates ranked suboptimal solutions, each representing an optimum solution containing a specific intron-exon junction.
• Error-model Optimization: Can be optimized for sequences with known levels of sequencing errors and handles substitution and frame-shift errors.

Scientific Applications:

• Human Gene Benchmarking: Tested on extensive collections of human genes to demonstrate robustness.
• Generalization to Novel Sequences: Effective on sequences not used during training, achieving a correlation coefficient of 0.89 for exon nucleotide prediction.
• G + C-rich Gene Analysis: Achieves a correlation coefficient of 0.94 for exon nucleotide prediction on G + C-rich genes.
• Robustness to Sequencing Errors: Quantitatively assessed for robustness against common sequencing errors and remains reliable on error-prone datasets.

Methodology:

Combines content statistics analysis, neural network weighting (approximating log-likelihoods for subintervals, including first/internal/last exons), and dynamic programming optimization to select exon–intron configurations and generate ranked suboptimal solutions, with provisions for handling substitution and frame-shift errors and tuning to known sequencing error levels.