Target Prediction Algorithms and Bioinformatics Resources for miRNA Studies - MicroRNAs as Tools in Biopharmaceutical Production - page 44

Fig. 3.2 Western blot analysis
of catalase and stathmin 48
and 96 hrs after miR7
transfection. Both proteins
were identified as potential
targets following analysis
with several in-silico target
prediction methods. (Adapted
from Meleady et al. 2012 )

through overlap with several algorithm predictions in mouse and rat. Shown below
(Fig. 3.2 ) is western blot analysis of these proteins following transfection with miR7
and the negative control, where the expression levels of both proteins are reduced
in comparison to cell with negative control cells at 48 and 96 h prioritising these
proteins for further experimental confirmation.
Expression profiling can aid in prioritising candidates for follow-up, however
expression level alone of a gene/protein changing cannot be directly attributed to
post-transcriptional regulation by a particular miRNA. Emergent techniques such as
HTS-CLIP which identify miRNA-mRNA interaction on a global scale (Chi et al.
2009 ) by direct analyses of sequences bound to the Ago proteins will increase the
number of validated direct targets in the future allowing us to gain a much clearer
picture of the performance of miRNA target prediction algorithm accuracy.

3.7 Conclusions

Target prediction algorithms will play a greater role in understanding CHO regu-
lations as we attempt to understand the impact of miRNA regulation on industrial
cell culture. The complexity of animal miRNA interactions has led to a proliferation
of computational techniques designed to predict which transcripts are regulated by
a particular miRNA. In this chapter a subset of the most widely used algorithms
and data repositories are described. Traditional algorithms follow multiple general
miRNA and mRNA sequence based rules, while emerging techniques utilise previ-
ously identified and experimental data to predict targets. The performance of these
target prediction algorithms has not been completely quantified as of yet due to the
lack of experimentally validated targets in the literature. There is no universal algo-
rithm for all target types; a general set of best practice guidelines have been suggested
and it is recommended to utilise several streams of evidence including algorithms
that interrogate the seed region (alignment, conservation, the presence of multiple
site etc.) and thermodynamic calculations and where possible, data from expression
profiling studies should be incorporated to filter results.