Towards computational prediction

of microRNA function and activity

 Abstract
• While it has been established that microRNAs(miRNAs)
play key roles throughout development and are
dysregulated in many human pathologies, the specific
processes and pathways regulated by individual miRNAs
are mostly unknown.
• Here, we use computational target predictions in order
to automatically infer the processes affected by human
miRNAs.
• Our approach improves upon standard statistical tools
by addressing specific characteristics of miRNA
regulation.
 Abstract
• Our analysis is based on a novel compendium of
experimentally verified miRNA-pathway and miRNA-
process associations that we constructed, which can be
a useful resource by itself.
• Our method also predicts novel miRNA-regulated
pathways, refines the annotation of miRNAs for which
only crude functions are known, and assigns differential
functions to miRNAs with closely related sequences.
• Applying our approach to groups of co-expressed genes
allows us to identify miRNAs and genomic miRNA
clusters with functional importance in specific stages of
early human development.
 Introduction
• miRNA: 19 – 25nt, non-coding RNAs that can reduce the
abundance and translational efficiency of mRNAs, and
play a major role in regulatory networks by the binding
of miRNAs to the 3’ UTRs of their targets.
• Reasons of limited success : redundant miRNAs and
compensatory effects in downstream signaling pathways
• If the targets of a specific miRNA are enriched with
genes annotated with some biological process or
pathway, it is reasonable to infer that the miRNA is
involved in the same process.
• Current method: HG test, log-likelihood test
• None of them was systematically tested for its ability to
recover known miRNA functions.
• They treat equally all predicted miRNA targets.
• Existing methods do not take into account the very
uneven distribution of 3’ UTR lengths.
 FAME
(functional assignment of miRNAs via enrichment)
• A new permutation-based statistical method that tests
for over- or under-representation of miRNA targets in a
designated set of target genes.
• FAME utilizes weights for miRNA-target pairs, accounts
for the number of miRNAs regulating each target, and
can be used for analysis of any group of miRNAs.
a. direct inference of miRNA function using sets of genes
sharing a common annotation.
b. indirect inference of miRNA function using matched
miRNA/mRNA expression data.
c. prediction of function for genomic clusters of miRNAs
 Materials
• miRNA target predictions: TargetScan
• 3’ UTR sequences: TargetScan
• miRNA and mRNA expression data
• co-expression clusters
• genomic clusters of miRNAs
• GO annotations
• KEGG pathway
 Method

𝑤 = 1 + 𝑘 ⋅ 𝑎 ⋅ 𝑏 ,𝑎 = 5𝑏 = 3
k - relative rank of the highest ranking conserved target site of m in t
a - discrete values w1. . .wa
b - control the relative contribution of the context scores to the
enrichment/depletion significance
 The context++ score (CS) for a specific site is
the sum of the contribution of 14 features
• site type • site8C*
• supplementary pairing • site8G*
• local AU • 3' UTR length*
• minimum distance • SA*
• sRNA1A* • ORF length*
• sRNA1C* • ORF 8mer count*
• sRNA1G* • 3' UTR offset 6mer count*
• sRNA8A* • TA (target site abundance)
• sRNA8C* • SPS (seed-pairing stability)
• sRNA8G* • PCT (probability of
• site8A* conserved targeting)*
 Results
gold standard
Overlap between
Overlap between predicted targets
TargetScan5 GO biological predicted targets and KEGG
family process term KEGG pathway Validated Targets PMIDs and GO term pathway
let-7/98
cell cycle Cell cycle CDK6; CDC25A 17699775 41 10
Toll-like receptor
let-7/98 immune response signaling pathway TLR4 19699171 29 6
miR-1/206 glucose
metabolism - IGF-1 18801338 9
miR-1/206 muscle
development - HCN2;HCN4 18458081 21
miR-1/206 regulation of 17062625,188013
muscle GJA1; IGF1; 38,16380711,159
development - HDAC4; HAND2 51802 1
miR-1/206 IGF1;
regulation of HSP60;HSP70; 17715156,159518
apoptosis Apoptosis HAND2 02 35 1
miR-106/302 regulation of cell
cycle Cell cycle E2F1 18328430 27 9
miR-124/506 PTBP1; CTDSP1;
neuron LAMC1; ITGB1; 17679093,173444
differentiation - SOX9 15 38
miR-125/351
apoptosis - TP54 19293287 43
miR-125/351
neurogenesis - LIN28 16227573 21
 Results
Direct prediction of miRNA functions-KEGG
 0% 10% 20% 30% 40% 50% 60% 70% 80% 90%100%

Results
mir-17-5p/20/93.mr/106/519.d: regulation of cell …
let-7/98: immune response
mir-106/302: regulation of cell cycle
mir-130/301: angiogenesis
Direct prediction of miRNA functions-GO
mir-1/206: glucose metabolism
mir-9: neuron development
mir-1/206: muscle development
mir-29abc: DNA modification
mir-155: B cell differentiation
mir-15/16/195/424/497: regulation of cell cycle
mir-125/351: apoptosis
mir-221/222: hemopoiesis
mir-1/206: regulation of apoptosis
mir-141/200a: epithelial cell differentiation
mir-17-5p/20/93.mr/106/519.d: regulation of …
mir-146: immune response
FAME
mir-192/215: regulation of cell cycle
mir-205: epithelial cell differentiation HG
mir-9: regulation of insulin secretion LLR
mir-133: regulation of apoptosis
mir-34a/34b-5p/34c/34c-5p/449/449abc/699: …
mir-125/351: neurogenesis
mir-29abc: regulation of apoptosis
let-7/98: cell cycle
mir-15/16/195/424/497: regulation of apoptosis
mir-124/506: neuron differentiation
mir-181: T cell differentiation
mir-200bc/429: epithelial cell differentiation
mir-21/590-5p: regulation of apoptosis
mir-133: muscle cell differentiation
mir-7/7ab: transmembrane receptor protein …
mir-133: muscle development
mir-21/590-5p: protein kinase cascade
mir-24: hemopoiesis
mir-34a/34b-5p/34c/34c-5p/449/449abc/699: …
mir-15/16/195/424/497: leukocyte differentiation
 Results
Using matched miRNA and mRNA expression to detection of
miRNA regulation
 Results
Prediction of function for genomic clusters of miRNAs
8 7
 Results
Reduce the 3’UTR length bias
 Discussion
• Limited accuracy of miRNA targets prediction methods
• Length and composition of 3’ UTR
• Limitations in the existing systems for functional
annotations
• The fact that most miRNAs have only a limited effect on
the expression levels of their target
• Limited by the quality of the target prediction
algorithms: conservation
• Difficult to predict miRNA target in coding sequence
FAME