Target Prediction Algorithms and Bioinformatics Resources for miRNA Studies - MicroRNAs as Tools in Biopharmaceutical Production - page 34

(John et al. 2004 ) and C. elegans —20 kcal/mol (Watanabe et al. 2006 ). Kertesz et al
have extended the principle of thermodynamic feasibility through the calculation of
the free energy of target accessibility ( G open ) governing the “opening” of mRNA
structure (target site along with upstream and downstream flanking sequences) to
facilitate the binding of miRNA to target (Kertesz et al. 2007 ).

3.3 miRNA Target Prediction Algorithms

3.3.1 miRanda

The miRanda algorithm was originally developed for target prediction in D.
melanogaster (Enright et al. 2003 ) and has now been extended to include human,
mouse, rat and C. elegans (Betel et al. 2008 ; John et al. 2004 ). miRanda initially
assesses the complementarity of an entire miRNA to each mRNA 3 UTR via an
extension of the classical Smith-Waterman alignment method. miRanda requires an
almost perfect match within the seed region, although wobble is permitted and can
be compensated by an extra match at the 3 end of the miRNA.
A sequence position dependent “weighted” alignment score (S) is calculated for
each potential mRNA target. Matches in the seed region of the miRNA are assigned
a higher weighting. The algorithm penalises G-U wobbles, gaps, insertions and
deletions. Scoring also takes into account the proximity of the target site to the
miRNA 5 end. The best scoring alignments are filtered on thermodynamic feasibility
by calculating the G duplex of the putative miRNA-mRNA interaction using the
Vienna algorithm (Wuchty et al. 1999 ). miRanda retains only interactions when
target sequences are above an evolutionary conservation threshold (e.g. PhastCons
score > 0.57) and the target site reside at approximately the same sequence position
(Betel et al. 2008 ; Enright et al. 2003 ) .

3.3.2 PicTar

The PicTar (probabilistic identification of target sites) (Krek et al. 2005 ) algo-
rithm differs from other target prediction strategies in that target sites for multiple
miRNAs as well as single miRNAs can be identified. PicTar, inspired by the observa-
tion of coexpression between miRNAs and the presence of multiple target sites within
3 UTRs (Hobert, 2004 ), attempts to reduce false positives through the identification
of common targets of coexpressed miRNAs. Target identification begins with a set of
miRNAs along with multiple orthologous 3 UTR sequence alignments and followed
by identification of perfect complementarity with the seed region or disruption of
complementarity by a single base difference (bulge, mismatch or wobble) at most.
These matches within regions of the 3 UTR termed “nuclei” are subjected to ther-
modynamic filtering of the resulting duplexes. A free energy threshold is applied,