Choosing the Best Gene Predictions with GeneValidator

Abstract

GeneValidator is a tool for determining whether the characteristics of newly predicted protein-coding genes are consistent with those of similar sequences in public databases. For this, it runs up to seven comparisons per gene. Results are shown in an HTML report containing summary statistics and graphical visualizations that aim to be useful for curators. Results are also presented in CSV and JSON formats for automated follow-up analysis.

Here, we describe common usage scenarios of GeneValidator that use the JSON output results together with standard UNIX tools. We demonstrate how GeneValidator’s textual output can be used to filter and subset large gene sets effectively. First, we explain how low-scoring gene models can be identified and extracted for manual curation—for example, as input for genome browsers or gene annotation tools. Second, we show how GeneValidator’s HTML report can be regenerated from a filtered subset of GeneValidator’s JSON output. Subsequently, we demonstrate how GeneValidator’s GUI can be used to complement manual curation efforts. Additionally, we explain how GeneValidator can be used to merge information from multiple annotations by automatically selecting the higher-scoring gene model at each common gene locus. Finally, we show how GeneValidator analyses can be optimized when using large BLAST databases.