| Gene hunts. Genes,
the regions that actually code for proteins, constitute
only a small fraction, perhaps 10%, of the human genome.
Thus, even with sequence in hand, finding the genes is
yet another daunting step away. One tool developed to
help in the hunt is GRAIL, a computer program developed
at Oak Ridge that uses heuristics based on existing data,
together with artificial neural networks, to identify
likely genes. Coding and noncoding regions of the genome
differ in many subtle respects -- for example, the
frequency with which certain short sequences appear.
Further, particular landmarks are known to characterize
the boundaries of many genes. In the example shown here,
GRAIL has searched for likely genes in both strands of a
3583-base-pair sequence. The results are shown at the
upper left. The upper white trace indicates five possible
exons (coding regions within a single gene) in one
strand, whereas the lower white trace suggests two
possible exons in the other strand. However, the lower
trace scores worse on other tests, leading to a candidate
set of exons shown by the five green rectangles. By
refining this set further, GRAIL then produces the final
gene model shown in light blue. The lower part of the
figure zeros in on the end of the candidate exon outlined
in yellow, thus providing a detailed look at one of the
differences between the preliminary and final models. The
sequence is shown in violet, together with the amino
acids it codes for, in yellow. The preliminary model thus
begins with the sequence GTCGCA. . . , which codes for
the amino acids valine and alanine. In fact, though,
almost all genes begin with the amino acid methionine, a
feature of the final gene model. At the upper right,
GRAIL displays the results of a database search for
sequences similar to the final five-exon gene model.
Close matches were found among species as diverse as
soybean and the nematode Caenorhabditis elegans. |
|
