mGene: accurate SVM-based gene finding with an application to nematode genomes

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Accurate gene prediction is crucial for understanding eukaryotic genome function.
• Existing gene finders have limitations in comprehensively identifying all protein-coding regions.
• The nematode Caenorhabditis elegans serves as a model system for evaluating gene prediction tools.

Purpose of the Study:

• To develop and evaluate mGene, a highly accurate gene-prediction system for eukaryotic genomes.
• To compare mGene's performance against established gene finders like Fgenesh++ and Augustus.
• To assess the impact of mGene on improving existing genome annotations and discovering novel genes.

Main Methods:

• Integration of generalized hidden Markov models (gHMMs) with Support Vector Machines (SVMs) for gene prediction.
• Objective performance evaluation using the Caenorhabditis elegans genome dataset.
• Experimental validation of novel predicted genes using RT-PCR and sequencing.

Main Results:

• mGene demonstrated superior prediction performance across nucleotide, exon, and transcript levels in objective competitions.
• The system identified approximately 2200 potentially novel genes in the C. elegans genome, with 42% confirmed by RT-PCR.
• mGene outperformed other gene finders in 10 out of 12 evaluation criteria and provided the most accurate predictions for four nematode species.

Conclusions:

• mGene significantly enhances the accuracy of eukaryotic gene prediction and genome annotation.
• The system has the potential to substantially improve gene catalogs, even for well-annotated genomes.
• mGene's predictions are valuable for comparative genomics and identifying species-specific gene inventions.