Biomedical Engineering Reference
In-Depth Information
3.4.7
miRecords
The miRecords (Xiao et al.
2009
) database contains both validated and predicted an-
imal miRNA targets. Predicted targets stored within miRecords are acquired from a
number of target prediction algorithms including TargetScan, PicTar, PITA, RNAhy-
brid, DIANA-microT and RNA22 as well as some machine learning approaches.
Also included are manually curated entries for validated targets from several species
extracted from the literature. At present the database holds 2,286 interactions with
experimental evidence (677 arise from profiling studies). From a search result a user
can see the degree of overlap between predicted target algorithms along with the re-
sults from the literature search including experimental manipulation of the miRNA
(over or under-expression) and effect on the mRNA or protein level.
3.4.8
miR2disease
As its name suggests miR2disease (Jiang et al.
2009
) focuses on cataloguing those
miRNA with an impact on human disease. The database contains 3,273 entries for
349 miRNA disregulated in 163 diseases, users can search for miRNAs, genes or
a disease of interest. For example a search for stomach cancer returns a list of
miRNAs with differential expression for several studies along with their role within
the disorder (e.g. causal). Each search returns the method used to detect miRNA
expression level, and validated targets from within the study or TarBase and links to
predicted targets for the microRNA.
3.5
miRNA Prediction Algorithm Performance
Which algorithm is most likely to provide direct miRNA targets? At present the
question is difficult to answer accurately due to large disparity between the numbers
of confirmed targets in comparison to the number of predicted targets. For example,
one study (Sethupathy et al.
2006
) compared target predictions for 84 mammalian
targets with 32 miRNAs in the current version of Tarbase that were confirmed using
direct methods, e.g. 3
luciferase assays. The study suffers from an obvious lack of
experimentally confirmed targets given the breadth of miRNA regulation; however
it is interesting to compare the performance of the individual algorithms against this
set. To determine the false positive rate, a set of
20 targets confirmed to have
no interaction were used. These experimentally confirmed targets and non-targets
were compared against those predicted by TargetScan, TargetScanS, DIANA-microT,
miRanda and PicTar along with the unions and intersections of various combinations
of the outputs of these algorithms. Each set of predictions were compared in terms
of sensitivity (S) (probability of detecting a true interaction). Early algorithms [Tar-
getScan (S
∼
=
21 %), DIANA-microT (S
=
10 %)] were found to have low sensitivity,
Search WWH ::
Custom Search