MicroRNA Target Prediction Based upon Metastable RNA Conformations

Student thesis: Doctoral ThesisDoctor of Philosophy


MicroRNAs (miRNAs) play an important role in biomarker research.
Identifying their targets and inferring their functions have been of a
great importance to developing our understanding of many biological
processes and fundamental novel anti-cancer and viral therapies.
Since the discovery and validation of true miRNA-messengerRNA
(mRNA) bindings is a laborious and expensive process, computational
tools for the prediction of miRNA targets are essential in this research
area. Advanced tools of miRNA target prediction incorporate
knowledge about secondary structures of mRNA sequences, usually
the 3'UTR into the evaluation and assessment of putative miRNAmRNA
bindings. The default secondary RNA structure in most target
prediction tools of this type is the minimum free energy conformation
or a representative of the ensemble of all possible RNA structures. A
key indicator of putative miRNA-mRNA bindings is the energy required
to open base pairs that are present in the potential binding
site within the conformation. However, mRNAs as well as miRNAs
are present in a single cell in multiple copies, where the number of
copies may range from several tens up to several hundreds of copies,
each of them transcribed from DNA at different points of time and
therefore, potentially, being present in different folding stages, most
likely in metastable conformations. In this thesis we have addressed
the problem of miRNA bindings to metastable RNA secondary structures
in the context of Single Nucleotide Polymorphisms (SNPs). To
this end, we first searched the recent literature for disease-related
triples [mRNA/3'UTR; SNP; miRNA] that have been analysed by
methods including PCR and/or luciferase reporter assays. We next
compared results of two major computational approaches to miRNA
target ranking prediction: conservation feature using TargetScan tool
and target site accessibility feature using PITA and STarMir tools. We
showed that site accessibility may be a better ranking criterion. We
then studied the problem of miRNA bindings to metastable secondary
structures in the context of SNPs and mRNA concentration levels
i.e. whether features of miRNA bindings to metastable conformations
could provide additional information supporting the differences
in expression levels of the two sequences defined by a SNP. We showed
that among the different parameters we introduced and analyzed, we
found that three of them, related to the average depth and average
opening energy of metastable conformations, may provide supporting
information for a stronger separation between miRNA bindings to the
two alleles defined by a given SNP. These findings were a trigger to
devise a novel target prediction tool that incorporates metastable secondary
structures with low energy levels into predictions. We present,
RNAStrucTar, a miRNA target prediction tool that analyses putative
mRNA binding sites within 3'UTR secondary structures representing
metastable conformations. The rst stage consists of generating
conformations that can be classified as deep local minima. The second
stage incorporates duplex structure prediction through sequence
alignment and energy computation. Target site accessibility related
to different sets of metastable conformations is also taken into account.
An overall interaction score computed from multiple binding
sites is returned. The approach is discussed in the context of SNPs
where our manually curated [mRNA;SNP;miRNA] dataset is utilised.
RNAStrucTar predictions are in favour of the allele with the stronger
miRNA binding stated in the underlying literature in 22 instances,
while the resulting scores are indifferent in ten cases. For the two
other cases (HTR3E and FGF20), the score is in favour of the weaker
allele. In this respect, RNAStrucTar results are better than PITA and
STarMir, with a positive prediction for RAD51 and MSLN (STarMir
favours the weaker allele).
Date of Award2017
Original languageEnglish
Awarding Institution
  • King's College London
SupervisorKathleen Steinhofel (Supervisor) & Sophia Tsoka (Supervisor)

Cite this