Biology Reference
In-Depth Information
on the information present in these genes, as opposed to also using
information that has so far emerged about the miRNA-target interac-
tion. Second, there are methods designed to predict miRNAs in a given
genome and methods that are directly or indirectly based on evolu-
tionary selection of miRNA precursors. Ideally, if we had an accurate
description of Drosha and Dicer substrates, one would expect to be
able to recognize them in individual genomes. Because such an accu-
rate description is lacking, most miRNA gene prediction methods aim
to identify miRNAs that are conserved in some number of species, or
use the property that the sequences of miRNA precursors were selected
during evolution (e.g. to stably fold into stem-loop structures) while
random stem-loops were not. Third, miRNA gene prediction methods
differ, of course, in the precise features that they use to distinguish
between miRNA precursors and other types of stem-loops. Finally,
these features are either used in filters that discard unlikely miRNA
precursors, or used to compute scores indicative of the likelihood
with which a stem-loop will be recognized and processed as a miRNA
precursor.
One of the first methods for miRNA gene prediction, miRseeker, was
developed by Lai
et al
.,
92
who focused on fruit fly miRNAs. The 24 fly
miRNAs that were known at the time were used to gauge the quality of
the prediction method. The procedure was to first identify intronic and
intergenic regions in the
D. melanogaster
genome that aligned well
(regions of 100 nucleotides with no more than 13% gaps and 15%
mismatches) to the genome of
D. pseudoobscura
. These regions were sub-
mitted to RNA secondary structure prediction, and stem-loops with at
least 23 base pairs and a free energy of folding
23 kcal/mol were
scored with a model that penalized internal loops of increasing size, par-
ticularly if they were asymmetrical. The orthologous regions of the
D. melanogaster
top candidates in
D. pseudoobscura
were then evaluated
in a similar fashion. Finally, the 600 candidates with the best average
score in the two genomes were submitted to a filter of their pattern of
evolutionary conservation. This was based on the analysis of the 24 reference
miRNAs, which indicated that mutations are least tolerated in the region
of the mature miRNA, followed by the complementary strand and finally
by the loop region. In the end, the authors suggested that about
∆
G
≤−
