Biology Reference
In-Depth Information
piecemeal way was obtained using the RNAeval
128
algorithm. The score
of a 3
′
UTR was computed as
Z
=∑
n
k
1
e
−
G
k
/
T
, where
n
is the number of
=
seed matches in the 3
UTR,
G
k
is the energy of miRNA:target site inter-
action for site
k
, and
T
is a free parameter. Predictions were obtained by
setting a cut-off on
Z
(
Z
i
′
≥
Z
C
, with
Z
i
being the score of the 3
′
UTR in
species
i
) and on
R
, the rank of the prediction (
R
i
≤
R
C
, with
R
i
being
the rank of the prediction in species
i
).
The third method proposed for miRNA target prediction was
miRanda,
124
initially applied to
D. melanogaster
target prediction. Unlike
TargetScan, which used an RNA secondary structure prediction program
based on energy minimization to predict hybrids between miRNAs and
target sites,
110
miRanda started from the premise that, in the context of
the RISC complex, the parameters of RNA-RNA interactions will
change. Therefore, miRanda used user-defined scores as opposed to
energetic parameters to predict hybrids (the scores of A-U, G-C, and
G-U interactions were 5, 5, and 2, respectively), and individual positions
in the miRNA were weighted differently in their contribution to the total
score. Mismatches were penalized with
−
3, and gaps were scored using
an affine model.
Also initially applied to the miRNA target prediction in
D. melanogaster
was the algorithm of Rajewsky and Socci,
108
which
was based on the identification of good matches between the miRNA
5
end and putative target sites, followed by further analysis of the
overall hybridization energy between the miRNA and the target site.
These authors further made the argument that the 5
′
end of the
miRNA implements an important kinetic component of miRNA-
mRNA interaction.
Since their initial inception, these target prediction methods have
changed (and converged) substantially.
104,125,129,130
In particular, evolu-
tionary conservation and close-to-perfect complementarity to the 5
′
end
of miRNAs have now been incorporated in virtually all target prediction
programs, while much less emphasis is currently placed on the energy of
miRNA-target site hybridization.
Given that miRNA-mRNA hybridization was believed to play such
an important role in miRNA target recognition, and that none of
the then-available RNA secondary structure prediction programs directly
′
