Analysis of some conventional ab initio gene finders using human and mouse DNA sequences
An evaluation of the prediction accuracy of five ab initio gene prediction programs (that is, FGENESH, Genscan, HMMgene, GeneMark.hmm and FGENES) was conducted by the use of 110 human and mouse orthologous sequences. As expected, all programs presented different predictions with various ranges of accuracy. According to our results, FGENESH and Genscan generally had the maximum power to produce more reliable results in both nucleotide and exon levels than others. Although, both FGENES and GenMark.hmm predicted the highest number of exons (966 and 946 exons, respectively), when exon sensitivity (ESn), exon specificity (ESp) and (ESn+ESp/2) were considered, their overall accurate performance descended and was clustered in the lowest positions. It was also determined that all programs have lower power in predicting initial and terminal exons, as compared to internal exons, which suggested that such programs cannot accurately determine translational start sites (TSS) and translational stop codons (TSC) as internal exons, whose boundaries are highlighted by acceptor and donor sites. Apart from the species difference, it was finally recognized that the programs, FGENESH and GeneMark.hmm, presented much more sensitivity in detecting genes with low guanine-cytosine (GC) content.
Key words: Ab initio gene prediction programs, human, mouse, orthologous sequences.