Gene recognition in eukaryotic DNA by comparison of genomic sequences

Motivation: Sequencing of complete eukaryotic genomes and large syntenic fr agments of genomes makes it possible to apply genomic comparison for gene r ecognition. Results: This paper describes a spliced alignment algorithm that aligns can didate exon chains of two homologous genomic sequence fragments from differ ent species. The algorithm is implemented in Pro-Gen software. Unlike other algorithms, Pro-Gen does not assume conservation of the exon-intron struct ure. Amino acid sequences obtained by the formal translation of candidate e xons are aligned instead of nucleotide sequences, which allows for distant comparisons. The algorithm was tested on a sample of human-mammal (mouse), human-vertebrate (Xenopus) and human-invertebrate (Drosophila) gene pairs. Surprisingly, the best results, 97-98% correlation between the actual and p redicted genes, were obtained for more distant comparisons, whereas the cor relation on the human-mouse sample was only 93%. The latter value increases to 95% if conservation of the exon-intron structure is assumed. This is ca used by a large amount of sequence conservation in non-coding regions of th e human and mouse genes probably due to regulatory elements.