Borodovsky developed the first version of GeneMark in 1993. In 1995, this program was used to find genes in the first completely sequenced genomes of bacteria and archea. The research team then developed self-training versions of the gene finding program for prokaryotic (organisms that lack a cell nucleus) and eukaryotic (organisms that contain a cell nucleus) genomes in 2001 and 2005, respectively. Development of these programs has been supported by the National Institutes of Health.
Unlike other programs that require a pre-determined training set along with the genome sequence, GeneMark.hmm-ES (BP) only requires the genome sequence. The program is able to iteratively identify the correct algorithm parameters from the anonymous sequence. The program uses a probabilistic mathematical model called the Hidden Markov Model to pinpoint the boundaries between coding sequences (exons) and non-coding sequences (introns and intergenic regions).
Most introns start from the dinucleotide guanine-thymine (abbreviated GT) and end with the dinucleotide adenine-guanine (abbreviated AG). However, finding these dinucleotides is not sufficient to signal the presence of an intron. Several nucleotides that surround GT and AG are also important, but the similarity of the pattern is not deterministic. Locating the branch site which is nine nucleotides in length, almost always contains an adenine and is located 20-50 bases upstream of the acceptor site helps to accurately identify an intron.
An initial run of the program with a reduced model containing heuristically defined parameters breaks the sequence into coding and non-coding regions. With this information, the researchers apply machine-learning techniques to refine the parameters of the recognition algorithm with respect to the sp
|Contact: Abby Vogel|
Georgia Institute of Technology Research News