Identification of protein coding regions in genomic DNA

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Accurate identification of protein-coding gene structures within genomic DNA is crucial for understanding gene function and regulation.
• Existing methods may face challenges in precisely delineating introns and exons, especially in complex genomic regions.

Purpose of the Study:

• To develop and evaluate GeneParser, a novel computer program for identifying and determining the fine structure of protein genes in genomic DNA sequences.
• To assess the accuracy and robustness of GeneParser in predicting exon-intron boundaries and structures.

Main Methods:

• GeneParser scores DNA sequence subintervals for statistical indicators of introns and exons, and their boundaries.
• A neural network weights these scores to approximate the log-likelihood of subintervals representing specific exon types (first, internal, last).
• A dynamic programming algorithm identifies the optimal combination of introns and exons maximizing the likelihood function.

Main Results:

• GeneParser achieved a high correlation coefficient of 0.89 for exon nucleotide prediction on unseen human gene sequences.
• Performance was even higher (0.94 correlation coefficient) on a subset of G+C-rich genes.
• The system demonstrated robustness against substitution and frame-shift sequencing errors, with optimization strategies identified.

Conclusions:

• GeneParser is an effective computational tool for accurately predicting gene structures, including introns and exons, in genomic DNA.
• The program's performance, particularly its robustness to errors, makes it valuable for analyzing diverse genomic datasets.
• This method facilitates rapid generation of ranked suboptimal solutions, aiding in comprehensive gene structure analysis.