A modification to an "ace" gene prediction program by computer scientists at Washington University in St. Louis now enables scientists to predict the very beginnings of gene transcription start sites and where the first splice occurs thereby defining the first exon of the gene.
The modification to the gene prediction software TWINSCAN is called N-SCAN. Michael Brent, Ph.D. professor of computer science at Washington University in St. Louis, together with Samuel S. Gross, then an undergraduate at Washington University, and Randall H. Brown, Ph.D., a research scientist, report their results in the May 2005 issue of Genome Research. N-SCAN has proven to be the best program available at finding both the transcription start site (TSS) and the complete first exon in both the human and fruit fly genomes.
The addition of N-SCAN to TWINSCAN now provides genomics researchers the wherewithal to find and predict both the protein sequences produced by genes and their untranslated regions. Researchers in recent years have grown increasingly enthusiastic about the significance of untranslated regions. By understanding the functions of these regions, scientists expect to understand more about gene regulation - how genes get turned on and off, the ignition system of our DNA, if you will.
To make the proteins that are the basic micro-machines of life, a region of the genome is copied, or "transcribed," to form a molecule called messenger RNA (mRNA). Some segments of the mRNA are then discarded, and the retained segments are spliced together. Geneticists have traditionally assumed that transcription starts within a few hundred bases of the protein-coding region. However, for nearly 40 percent of known human genes, transcription starts long before the beginning of the protein- coding region. Most of this extra-long untranslated region is then discarded by splicing the 5' untranslated region (UTR). All present gene finding systems - except for N-SCAN - either ignore the UTR splice sites or incorrectly incorporate them into some protein-coding segment, making gene prediction a none-too-sure industry.
"We've found that when we add the spliced untranslated regions to our system, we not only get good predictions for UTRs but also improved predictions of the protein-coding region of the gene. By correctly identifying UTRs, we can avoid labeling them incorrectly as part of the protein-coding region," said Brent, who, with various colleagues, developed both TWINSCAN and N-SCAN. "It's important to know these two areas. Some of the signals that regulate transcription reside right near the transcription site. There is a huge amount of biology to be discovered there, and the appreciation of this area is growing daily."
While genomics researchers 15 years ago paid little attention to parts of the genome outside the coding regions, they have discovered some strange functions in UTR that have provoked second and third thoughts.
For instance, it recently was discovered that huntingtin, a gene associated with Huntington's disease, has a second protein segment encoded upstream of the main one. This protein in the so-called untranslated region is involved in regulating the gene. Running the modified TWINSCAN, on both the human and fruit fly genomes, Brent and colleagues predicted about 25,000 transcription-start sites, compared with a known 6,000.
"In the human genome, we found many extra exons on genes that were already known, or in some cases, spliced UTRs on genes that weren't even known to exist before," Brent said.
The system takes advantage of the scarcity of the CG sequence, finding so-called CpG "islands" known to be more common near the transcription-start site. It also has a knack for recognizing sequences that indicate splice sites. Over the past two years, TWINSCAN has been finding and predicting genes in numerous genomes that other gene prediction systems have missed. The addition of N-SCAN to the handy system - it scans two genomes simultaneously, with potential to scan three or more - strengthens it for predicting both coding and non-coding DNA.
"Like any multiple choice question, if you can learn something about one of the choices, it helps you with the other one," Brent said. "By making this integrated model that looks for both kinds of exons in both parts of the gene, we're able to convert a blind guessing game to a multiple choice question - is it a UTR exon or a protein-coding exon? These kinds of questions are easier to answer now."