What is miRNA pharmacogenomics?
As the name suggests, microRNA (miRNA) pharmacogenomics combines the fields of miRNAs and pharmacogenomics.
miRNAs are short (~22 nucleotide) non-coding RNAs that down-regulate gene expression by binding to 3’ UTRs of mRNAs leading to translational repression and degradation of the mRNA.
Pharmacogenomics is the study of individual differences in genome and transcriptome composition and their effect on drug safety and efficacy. miRNAs have recently gained attention within the field of pharmacogenomics, since miRNAs can affect the expression of drug associated genes eventually affecting drug function itself.
What is a [miRNA - Gene - Drug] set?
miRNAs can affect drug function by down-regulating genes that are
important for drug efficacy and/or toxicity.
Pharmaco-miR contains information on predicted or tested miRNA targets in pharmacogenomics and their related genes as well as drugs that may be affected by these miRNA-gene interactions.
We refer to these sets as [miRNA - gene - drug] sets, to underline the causal nature of the interactions.
How are searches performed in Pharmaco-miR?
To search Pharmaco-miR, enter a miRNA name, gene symbol and/or generic drug name in the search box:
- Names and symbols can be entered in random order, but must occur on separate lines
- miRNAs are entered in the form of miR-XX or let-7X
- Gene symbols would look like "BCL2" or "CYP3A4"
- Drugs names would look like "ibuprofen" or "cisplatin"
You can also insert objects by using the buttons on the right and clicking on the desirable name, or upload a file from your computer in which the entries are in seperate lines.
How are miRNA targets identified in Pharmaco-miR?
As default, Pharmaco-miR searches experimentally validated
miRNA targets as annotated in miRecords
Furthermore, Pharmaco-miR uses three different miRNA target prediction algorithms: TargetScan , miRanda and PITA .
The default search options include the verified databases: VERSE, miRecords, and mirTarBase. Since a large proportion of miRNA targets are still thought to be unvalidated, TargetScan is also included in the default search options, and miRanda and PITA can be added in the search parameters.
As a default, Pharmaco-miR searches only for sites identified in TargetScan as conserved between species, which should decrease the number of false positive miRNA target site predictions.
Which genes are included in Pharmaco-miR?
Genes appear in Pharmaco-miR by their gene symbol. The database
includes all genes that are annotated as associated with drug function
in the Pharmacogenomics
Knowledge Base .
Some genes are not predicted as miRNA targets and will therefore not occur in any [ miRNA - gene - drug ] sets.
If searching for sets including a specific gene does not yield any output, the gene might therefore:
- be spelled incorrectly
- not be associated with any drug
- not contain any miRNA targets under the given search parameters
Which drugs are included in Pharmaco-miR?
Drugs appear in Pharmaco-miR by their generic names. Furthermore, drug classes (as listed in the drug.tsv file on PharmGKB ) are included.
How are Gene - Drug associations identified in Pharmaco-miR?
Gene importance for drug function is based on literature annotation,
as performed by the Pharmacogenomics Knowledge Base (downloadable from
the relationships.tsv file on PharmGKB
The exact nature of the gene - drug interactions has not been determined and it may therefore be necessary to consult the specific papers to determine the associations and pathways (these associations derive from the pathways on the Pharmacogenomics Knowledge Base site, which describe the gene - drug interactions of a drug function).
Where can I download Pharmaco-miR VERSE database?
VERSE can be downloaded here .
How are TargetScan target site conservation and miRNA conservation levels defined?
miRNA conservation levels are defined as follows:
- Broadly conserved: Conserved across most vertebrates, usually to zebrafish
- Conserved: Conserved across most mammals, but usually not beyond placental mammals
- Poorly conserved: All others
Information on target site conservation can be found here and in the following paper (which also contains more info about miRNA conservation levels):
Most Mammalian mRNAs Are Conserved Targets of MicroRNAs; Robin C Friedman, Kyle Kai-How Farh, Christopher B Burge, David P Bartel. Genome Research, 19:92-105 (2009)
What is the difference between choosing ‘All associations and ‘Overlapping associations’ in the search options?:
The optimal type of output of a Pharmaco-miR search depends on the
scientific question asked.
The biggest amount of information is achieved when all [ miRNA - gene - drug ] pharmacogenomic sets that include the search terms (assuming there are several) are shown. This corresponds to an ‘All associations’ search.
However, depending on the search entries, the number of output pharmacogenomic sets may number in the hundreds, and at times, it may be more relevant to examine if all the entered search terms are connected by one or a few miRNAs, genes or drugs.
For instance, if 4 genes all associated with a drug are found deregulated in an expression study, it may be relevant to discover, whether all of the 4 genes are targeted by the same miRNA. To do this, choose ‘Overlapping associations’ as the output type.
In this case, the output only includes miRNA pharmacogenomic sets that may explain variability in all the entered search terms at the same time.
In the example on the left, if genes 1-3 (shown in red) are entered as search terms, choosing ‘All associations’ will include sets with miR-A, miR-B and miR-C, since each of these miRNAs targets at least one of the genes (our search terms).
Choosing ‘Overlapping associations’ will include only sets with miR-B, since it is the only miRNA which associates all three genes and can therefore explain the variations. In the output, miR-B will be written in bold letters to designate that this is the unit which associates all search terms.
Please note: when using ‘Overlapping associations’, it is
currently only possible to insert search terms of only one type
(miRNAs, genes or drugs).
This restriction does not apply on the ‘All associations’ feature.
What is the difference between the ‘Detailed associations’ and ‘Reduced associations’ tables?
Since [ miRNA - gene - drug ] sets often number in the hundreds for an
individual search, the search results page may be structured in two
different ways in pharmaco-miR.
‘Detailed associations’ lists contain the full miRNA pharmacogenomic sets, each consisting of a miRNA, a target gene and an associated drug.
However, to give a first impression of the data and limit the length of the output, ‘Reduced associations’ may also be chosen. In this case, the output miRNAs, genes and drugs will not be shown in sets, but will be unique and shown in three separate columns. In this way, a quick overview of which miRNAs, genes and drugs are present in the output data can be achieved.
What are my options for exporting search results?
To export search results, simply press the download button above the search results, and the entire output (not only the sets present in the current window) will be exported. The downloaded file is .csv formatted (comma delimited) and should automatically open in Microsoft Excel.
My output is very large. How should I proceed?
Depending on the input you choose, the Pharmaco-miR output may be
quite large, potentially numbering thousands of miRNA pharmacogenomic
In this case, there are several ways to manage the data:
- Choose a lower number of target databases. As a default, Pharmaco-miR searches miRTarBase, miRecords (both with experimentally tested sites), and TargetScan (computational predictions). Well studied genes and miRNAs will often have several entries which have been experimentally verified and therefore are more likely to be relevant in a pharmacogenomic context. In these cases, searching only miRecords and miRTarBase will yield a shorter list of high confidence sets.
- Show output as a ‘Reduced Associations’ table. While the information in a reduced associations table is less detailed, it allows a quick overview of the data to check for potentially interesting observation. One can then revert to the detailed associations view to identify the full sets of interest.
- Export tables to Excel. Clicking the ‘Download’ button will open the table in Microsoft Excel. From here, tables can be searched in detail using the Excel table features.
- Use links. While the varied databases utilized in Pharmaco-miR makes it impossible to score the identified targets, clicking on an entry will open a list of links. These can then be followed to the relevant databases and reveal a detailed view of the entries. In this way, the strength of the different sets can be assessed.
How do I distinguish between more or less potent miRNA pharmacogenomic sets in my output?
Pharmaco-miR output consists of miRNA pharmacogenomic sets, sometimes in great numbers. Deciding on which of these are relevant in a specific context is no trivial task. It depends on the scientific question asked and on the validity or strength of the Pharmaco-miR predictions, including the miR::gene pairs and gene-drug associations. For the miRNA::gene links, how likely are the predicted sites to be functional and how big an impact will the miRNA have on gene expression? And for gene-drug associations, what impact will lowered gene expression have on drug function?
An initial filtration of miRNA-pharmacogenomic sets, to start out with a relatively conservative set of predictions, can be achieved already when choosing the search settings. All miRNA prediction databases are represented with several permutations of the database, with more or less likely target predictions, based on conservation (as a proxy for function) and/or sequence and structural features known to promote miRNA binding. These can be toggled on and off on the main search page. Details on the different algorithms can be found in the 'About' page and by pressing the question mark next to the database names on the search page.
The miRNA predictors included in Pharmaco-miR estimate miRNA efficacy using different algorithms based on different features, and thus scores the miRNA targets differently. It is therefore difficult to compare the target site prediction directly. However, Pharmaco-miR provides the relevant scores for the databases predicting the miRNA::gene pairs. Columns with the prediction score for the different databases can be toggled on and off in the search results and are exported to excel together with the miRNA pharmacogenomic sets for further processing.
The most conservative sets are achieved by only using miRNA targets that are already verified experimentally. Two of the databases included in Pharmaco-miR, miRTarBase and miRecords, contain such target sites curated from literature. When targets are registered in one of these databases, a link to the relevant papers will also be available. However, only a small fraction of the expected functional miRNA targets have to date been tested experimentally. Although including only targets from miRTarBase and miRecords will yield high confidence sets, it will therefore likely also miss a lot of potentially important predictions. Conserved miRNA targets predicted by TargetScan are therefore also included in the default search parameters.
Gene-drug associations derive from Pharmacogenomics Knowledge Base and the database developed specifically for Pharmaco-miR containing complete miRNA pharmacogenomic sets. All of these associations are based on literature annotations and links to the relevant papers are available in Pharmacor-miR. Furthermore, as a proxy for the importance of the gene-drug association, a count of the number of papers describing an association is displayed, based on the assumption, that the more essential a certain gene is for the function of a drug, the more times the association will have been described in literature.
Which database files are used in Pharmaco-miR?
miRecords and mirTarBase are collections of experimentally verified miRNA targets curated from literature. Data from these two databases are included in Pharmaco-miR miRNAs::gene pairs together with information on which paper describes the targeting. Pharmaco-miR contains links to these papers and, as a proxy for the amount of support for the target, a count of the number of papers describing the miRNA::gene interaction is included.
TargetScan was the first miRNA target database published (Lewis et al. 2003). It was originally launched as part of the first major target search, revealing the importance of the seed region (Lewis et al. 2003), but has been regularly updated and improved (Grimson et al. 2007; Friedman et al. 2009). The database includes all potential target sites with minimum a 6mer seed match, but does not allow for mismatches in the seed sequence. It permits stringent filtering based both on target conservation and conservation of the miRNA family. Target sites can furthermore be evaluated based on seed type and context contribution (a score summarizing the contribution of several parameters known to impact the efficacy of a target) as well as depth of evolutionary conservation (Lewis et al. 2005; Grimson et al. 2007; Friedman et al. 2009).
Pharmaco miR allows several different filtering options when including TargetScan predictions in searches. miRNAs can be either broadly conserved, conserved or non-conserved (as defined by TargetScan and described on their homepage). Similarly, the miRNA targets can be conserved or non-conserved. Conserved miRNAs and targets generally have a more pronounced ability to inhibit gene expression and have more biologically relevant targets. Furthermore, each miRNA-gene pair extracted from TargetScan to Pharmaco-miR includes a context score, indicating the total predicted effect of the miRNA on the gene’s expression level.
TargetScan predictions are scored by a ‘context score’ which combines contributions from a wide range of features known to affect target site efficacy, including seed type, compensatory binding in the 3’ of the miRNA, AU content and position on the 3’UTR. Lower values indicate stronger downregulation of expression. Total context score for a miRNA::gene pair is calculated by adding scores for the individual target sites on the gene.
The miRanda algorithm (John et al. 2004) uses less stringent criteria than TargetScan for prediction of target sites. miRNA are aligned to mRNAs to identify complimentarity, but while seed match is weighed more strongly, G:U wobble base pairing and mismatches in the seed does not lead to exclusion from the set of predictions (Betel et al. 2008). Similarly, the algorithm considers conservation throughout the potential miRNA matching region, rather than only seed conservation. Targets are partly scored based on the free energy of the entire mRNA:miRNA heteroduplex. miRanda therefore includes de facto targets that may be missed by more stringent algorithms; however, this increase in sensitivity is likely accompanied by a loss of specificity.
In Pharmaco-miR, miRanda predictions can be searched independently for high and low scores. Similarly, conserved and nonconserved miRNAs can be toggled on and off independently in the search feature options.
Furthermore, miRanda targets are scored by their mirSVR score, generated by a machine learning algorithm trained on expression data from miRNA transfected cells. It incorporates a large range of features, both structural and sequence based. Thus, mirSVR scores for a miRNA::gene pair constitute an estimate of the miRNA’s effect on the gene expression level. These were determined by adding the mirSVR scores for the individual targets for a specific miRNA in the gene, as described in the mirSVR score documentation (Betel, 2010). As with the other target prediction scores, a lower (more negative) value indicates stronger miRNA::gene binding.
PITA (Kertesz et al. 2007) differs from the two above miRNA target prediction algorithms mainly by focusing not only the miRNA:mRNA match, but also structural features surrounding the putative target site and the energy cost to open and expose the target site to the miRNA loaded RISC. Full seed match is required for 6mer sites, while 7mer and 8mer matches are allowed one wobble G:U pair. To achieve more conservative predictions it is furthermore possible to include conservation as a parameter.
In Pharmaco-miR, PITA target predictions are included in two sets; Predictions with strong seed matches and high conservation (‘top targets’) and a set of target predictions which also include predictions with a lower level of support (‘all targets’).