Biomedical Engineering Reference
In-Depth Information
with more recent algorithms [(miRanda (S
=
49 %), TargetScanS (S
=
48 %) and
PicTar (S
48 %)] significantly increasing in sensitivity. Combining the results from
the five algorithms had the highest sensitivity (S
=
=
100 %) outperforming individual
algorithms and also the union of predictions.
Proteomic and transcriptomic profiling following miRNA transfection/knockout
have also been utilised to test prediction algorithms. Two studies compared predicted
outputs from miRBase Targets, miRanda, PicTar, PITA and TargetScan in terms of
those principles outlined in Sect. 3.2 of this chapter. In this comparison PicTar and
TargetScan had the highest prediction rate for those proteins repressed amongst
the algorithms (although
>
66 % of the predicted targets were not detected). The
authors noted that this was due to the utilisation of site conservation and perhaps
stringent pairing with the seed—analysis of perfect seed match against mismatched
seed reduced the benefit of conservation analysis. Ranking predictions based on total
context score (Grimson et al.
2007
) proved to be a particularly effective means to
identify proteins that were downregulated (Baek et al.
2008
). Further analysis of
the precision (P) (P
correct prediction/total prediction) and sensitivity of in silico
target predictions (Alexiou et al.
2009
) on the proteome (Selbach et al.
2008
) further
highlighted the effectiveness of TargetScan/TargetScanS (P
=
=
51 %/P
=
49 %) and
=
=
=
PicTar(P
49 %). In comparison, the miRanda (P
29 %), PITA (P
26 %) and
=
RNA22 (P
24 %) algorithms were found to have a low precision with high false
positive rates.
It is clear that there is no one target prediction algorithm detects all targets of a
miRNA. Current consensus when prioritising targets is to utilise multiple target pre-
diction algorithms (Table
3.1
) in combination with sequence and laboratory analysis.
The algorithms chosen should focus both on seed site conservation (e.g. TargetScan,
PicTar and Diana-microT) and thermodynamic calculations (PITA, RNA22). If pos-
sible global profiling tools such as proteomics or microarrays should be used to
compare expression patterns of the predicted targets. In addition the sequence fea-
tures of targets such as multiple sites in close proximity should be used to further
identify candidates (Table
3.2
).
3.6
Utilisation of miRNA Target Prediction in CHO Research
The CHO field is beginning to move beyond sequence and differential expression
analysis to determining miRNA function. Two recent studies have illustrated how
the guidelines listed above and integration of predictions with analytical data can be
applied for candidate selection. The first study investigated miRNA expression and
gene expression in parallel across the growth cycle (lag, exponential, stationary and
decline) in order to determine the effect of miRNA on CHO cell function and tran-
scriptome (Bort et al.
2011
). Clusters of genes with similar expression patterns were
compared to the target predictions for differentially expressed miRNAs from the 11
target prediction algorithms via the GeneSet2miRNA interface (Antonov et al.
2009
).
Only targets predicted by at least four algorithms were retained. Anti-correlation of
Search WWH ::
Custom Search