GeneLook: a novel ab initio gene identification system suitable for automated annotation of prokaryotic sequences

With the rapid increases in the amounts of sequence data for prokaryotic genomes, it has become important to develop systems for automated and accurate genome annotation. We present herein a novel ab initio gene identification system, GeneLook, that predicts protein-coding open reading frames (ORFs) with high sensitivity and specificity with no prior knowledge of the sequence composition. The system predicts protein-coding ORFs in two stages, seed ORF selection and main prediction. In the selection of reliable seed ORFs containing at least 200 codons, GeneLook predicts translation start sites and operon structures through searches for ribosome-binding sites and a novel operon prediction algorithm. The codon and nucleotide frequencies of seed ORFs are then used to determine values for two new coding-potential parameters for identification of protein-coding ORFs of at least 34 codons and for another parameter that improves the prediction accuracy for GC-rich genomes. In the main prediction, GeneLook uses these parameters to identify the most likely genes of a given minimal length. We assessed the performance of GeneLook with two indices, sensitivity and specificity that are defined as true positives (TP)/(TP+false negatives) and TP/(TP+false positives), respectively. This system predicted protein-coding ORFs for Escherichia coli and Bacillus subtilis with sensitivities of 96.5% and 96.2%, respectively, and specificities of 96.9% and 96.1%, respectively. The system also identified 94.1% of annotated genes of the Pseudomonas aeruginosa genome, which is GC-rich, with high specificity (97.2%). Furthermore, GeneLook identified protein-coding ORFs with high accuracy from a wide variety of prokaryotic genomes.