TimiRGeN: Time Incorporated miR-mRNA Generation of Networks
by Krutik Patel
1 Introduction
Time Incorporated miR-mRNA Generation of Networks (TimiRGeN) is aimed at researchers who wish to explore interactions in time series microRNA-mRNA expression data. This package integrates, functionally analyses and generates small networks for hypothesis generation.
To achieve data reduction without reducing biological signal, the TimiRGeN package utilises several published packages and employs their functions in a synergistic fashion for time series multi-omic analysis. The following packages have been built upon for several functions in the TimiRGeN package:
rWikiPathways [1], clusterProfiler [2], DOSE [3], biomaRt [4], RCy3 [5], Mfuzz [6], igraph [7].
TimiRGeN is very selective and only uses miR-mRNA interaction data from databases curated within the last 2 years. To reduce the number of false-positives, TimiRGeN also only uses predictive databases which use seed site specificity as their main input.
TargetScans[8], miRDB[9], miRTarBase[10].
TimiRGeN does have the capability to generate networks in R, however this package is uniquely open ended, as the output can be easily be exported to cytoscape [11] or pathvisio [12] for better visualisation options.
TimiRGeN solely uses wikipathways for functional pathway analysis, and is the first tool to allow this for time series data. Wikipathways is a user curated pathway database that contains 1000s of mechanistic signalling pathways from multiple species [13]. Furthermore, wikipathways works very well with pathvisio which is our recommended tool for GRN (gene regulatory network) design. Please read the TimiRGeN/inst/Pathvisio_GRN_guide.pdf for a step-by-step tutorial for our GRN creation process.
The TimiRGeN package has several options for miR-mRNA analysis. Currently the package can analyse human or mouse data, perform analysis of miR and mRNA data combined or separately, and can use entrez or ensembl gene IDs. This is because most wikipathways are annotated with either entrez IDs or ensembl gene IDs. This tool can be best used after differential expression (DE) analysis, and has potential to become a staple part of any miR-mRNA expression data study.
1.1 Install TimiRGeN
2 Combined miR-mRNA analysis
2.1 Example Mouse Kidney Fibrosis Dataset
In this section the combined method will be used to analyse a mouse kidney fibrosis data set. The mRNA data was published in Craciun et al (2016) [23] which was downloaded from GSE65267. The associated miR data was published in Pellegrini et al (2016) [24] and this was downloaded from GSE61328.
Notice the standard nomenclature used in the column names. Do follow the this standard for your own input data. The time point should come first and is followed by a .. The time point should consist of alphabetical characters followed by numerical characters e.g. D1, H6, TP3. After the . the column name should continue to display the specific result types from differential expression analysis.
Note. There should only be one . in each column name and no _ characters. Having more than one . or any _ characters will confuse some functions.
Note. There should be no NAs in your miR and mRNA data files.
2.2 Create MultiAssayExperiment object
TimiRGeN uses MultiAssayExperiment (MAE) to contain information. The dataframes and matrices will be stored as assays, S4 objects will be stored as Experiments and the lists will be stored as metadata.
If unfamiliar with MultiAssayExpriments please read through the vignette to understand how data can be accessed or go through the user guide which can be found on the MultiAssayExperiment bioconductor page [22].
2.3 Retrieve Gene IDs
MAE <- getIdsMirMouse(MAE = MAE, assay(MAE, 1))
MAE <- getIdsMrnaMouse(MAE = MAE, assay(MAE, 2), mirror = "useast")Using getIds functions to produce dataframes containing entrezgene and ensembl ID annotations for genes. This is useful for downstream analysis.
Many wikipathways use either entrezgene IDs or ensembl gene IDs for annotation. Having both formats available can be useful.
Due to the nature of miRs, many NAs may be found in the output of getIdsMir functions. Entrezgene IDs and ensemble IDs are insensitive to miRs with -3p and -5p strands. Therefore, adjusted entrezgene IDs and ensemble IDs are also created.
Note. For getIdsMrna functions, if a connection time out error occurs or if downloads are very slow, try to use other mirrors e.g. mirror = "useast".
2.4 Filter Out Non-significant Genes
MAE <- combineGenes(MAE = MAE, miR_data = assay(MAE, 1), assay(MAE,
2))
MAE <- genesList(MAE = MAE, method = "c", genetic_data = assay(MAE,
9), timeString = "D")
MAE <- significantVals(MAE = MAE, method = "c", geneList = metadata(MAE)[[1]],
maxVal = 0.05, stringVal = "adjPVal")mRNA and miR data can be combined using combineGenes function.
The genesList function will transform the large dataframe into multiple nested dataframes within a list. The data will be separated by the timeString parameter. In this example by D (days), because it was the non-numeric character before the . in the column names.
Significantly differentially expressed genes can be retrieved from each nested dataframe using the significantVals function. In this example. only genes which had an adjusted P value of less than 0.05 would remain in the list.
2.4.1 Find Significant Gene IDs
MAE <- addIds(MAE = MAE, method = "c", filtered_genelist = metadata(MAE)[[2]],
miR_IDs = assay(MAE, 3), mRNA_IDs = assay(MAE, 7))
MAE <- eNames(MAE = MAE, method = "c", gene_IDs = metadata(MAE)[[3]],
ID_Column = 4)Now entrezgene IDs or ensembl IDs which were created before can be integrated into the filtered dataframes of genes using addIds. In this example entrezgene IDs were added.
Lists of entrez IDs/ ensembl IDs can be extracted for further analysis using the eNames function.
2.5 Time Dependent Pathway Enrichment Method
Once we have a list of significant genes per time point we can put this through gene set enrichment analysis to find enriched pathways in each time point in the data. TimiRGeN uses wikipathways [13] for GSEA.
This is standard GSEA, here the enrichWiki function wraps around enrichment functions from DOSE and clusterProfiler [2,3] but applies these functions for time series analysis with wikipathways.
Note. Making multiple separate MAE objects makes it easier to work with all the generated data files.
MAE2 <- enrichWiki(MAE = MAE2, method = "c", ID_list = metadata(MAE)[[4]],
orgDB = org.Mm.eg.db, path_gene = assay(MAE2, 1), path_name = assay(MAE2,
2), ID = "ENTREZID", universe = assay(MAE2, 1)[[2]])path_gene and path_name can be found as output from the dloadGmt function.
For a more stringent check, a unique universe can be used e.g. all possible genes found in a microarray or all known genes expressed in a cell type.
2.5.1 Plot GSEA
To plot results from GSEA, the savePlots function can save all plots in the current working directory. Either bar plots or dot plots can be generated by using either quickbar or quickDot, and the plots can be saved to file in a variety of formats.
From here a wikipathway can be selected for miR-mRNA interaction analysis. If a pathway of interest is found, move onto part 2.7 of the vignette.
2.6 Temporal Pathway Clustering Method
TimiRGeN also offers a supervised soft clustering method using Mfuzz [6, 25]. This provides a global view of temporal patterns based on the common genes found between the each time point of the data and each pathway.
2.6.1 Create Percent Matrix
This method starts by importing all mouse wikipathways and associated gene IDs. The L symbol will give us entezgene IDs. Use the En symbol for ensembl IDs.
wikiMatrix will produce a matrix of common genes found between each time point (rows) and each pathway (columns). A final row representing the total number of genes in each pathway is added to the bottom of the matrix.
turnPercent will normalise the number of genes found in each pathway by converting to percentages. This will account for pathways of differing sizes.
2.6.2 Temporal Pathway Clustering
MAE2 <- createClusters(MAE = MAE2, method = "c", percentMatrix = assay(MAE2,
5), noClusters = 12, variance = 0.99)createClusters uses functions from Mfuzz [6] to separate the percentMatrix into a number of clusters based on temporal changes. In our example we created 12 clusters. Please refer to the Mfuzz source material for more information [25].
Note. The message “genes excluded” refers to number of pathways excluded in TimiRGeN.
A standard deviation plot is created to show the variance found in each pathway over the time course. A PCA plot can also be created, this will show how close each of the clusters are. Both plots are from Mfuzz functions [6].
2.6.3 Plot Clusters
Once we are happy about the number of clusters, the fuzzy clusters can be plotted. Standardized values representing abundance are on the y axis and the time points are on the x axis.
Each cluster represents a different temporal behaviour, and each cluster will consist of lines which will fit to varying degrees to each temporal behaviour. Colours of the lines will represent how well a particular pathway fits into a temporal behaviour.
Red lines fit very well, orange lines fit fairly well, yellow lines somewhat fit and purple lines do not fit well.
Note. Clear plots after quickFuzz, as there could be visual defects in future plots.
2.7 Take Selected Pathways Forward
MAE2 <- diffExpressRes(MAE = MAE2, df = assay(MAE, 1), dataType = "Log2FC",
genes_ID = assay(MAE, 3), name = "miR_log2fc")
MAE2 <- diffExpressRes(MAE = MAE2, df = assay(MAE, 2), dataType = "Log2FC",
genes_ID = assay(MAE, 7), name = "mRNA_log2fc")The diffExpressRes function will retrieve one specific result type from differential expression, along with either entrez gene IDs or ensembl gene IDs from a getIDs function. It is recommend to use a DE results type which represents abundance e.g. Log2FC, average expression or normalised expression values, because these values will be averaged.
Add a name to differentiate the results from diffExpressRes. Objects within MAEs cannot share the same name.
Note. Parameter df of diffExpressRes should not contain any NAs. Parameter genes_ID can contain NAs.
2.7.1 Find Common Genes in the Pathways of Interest and the Input mRNA Data
MAE2 <- reduceWiki(MAE = MAE2, path_data = assay(MAE2, 3), stringWiki = "Lung fibrosis")
MAE2 <- wikiMrna(MAE2, mRNA_express = assay(MAE2, 10), singleWiki = assay(MAE2,
11), stringWiki = "Lung fibrosis")Now we can identify which genes from a pathway of interest are in common with the input data. The Lung Fibrosis pathway is used in this example.
2.8 Create Correlation Matrix
MAE3 <- MultiAssayExperiment()
MAE3 <- mirMrnaInt(MAE = MAE3, miR_express = assay(MAE2, 9),
GenesofInterest = assay(MAE2, 12), maxInt = 5)mirMrnaInt creates a correlation matrix for every possible miR-mRNA interaction between the GenesofInterest created from wikiMrna and all the input miRs.
The maxInt parameter must be equal to the number of time points in the input data.
2.9 Download miR-mRNA Interaction Database Data
TargetScans v7.2, miRDB v6.0 and miRtaRBase v8.0 [8-10] can be used to determine which of the potential miR-mRNA interactions have been predicted/ functionally validated.
The following code will download the latest versions of the databases and format them for species specific miR-mRNA interaction mining.
For human analysis use "hsa" and org.Hs.eg.db.
2.10 Mine Predicted or Functionally Assessed miR-mRNA Interactions
MAE3 <- dataMiningMatrix(MAE = MAE3, corrTable = assay(MAE3,
1), targetscan = assay(MAE3, 2), mirdb = assay(MAE3, 3),
mirtarbase = assay(MAE3, 4))The dataMiningMatrix function will identify if any of the potential miR-mRNA interactions from the mirMrnaInt function are predicted/ functionally assessed, and display this as integers.
2.10.1 Filter Out miR-mRNA Interactions Based on Evidence
MAE3 <- matrixFilter(MAE = MAE3, miningMatrix = assay(MAE3, 5),
negativeOnly = TRUE, predictedOnly = FALSE, threshold = 2,
maxCor = 0.5)Potential interactions with no-little evidence can be filtered out using parameters on matrixFilter.
NegativeOnly being TRUE will only filter for interactions with a negative average correlation. PredictedOnly being TRUE will ignore interactions found on miRTarBase. THRESHOLD being 2 will filter for interactions that have been predicted/ validated by two or more databases. maxCor being 0.5 will remove any interactions that have an average correlation greater than 0.5.
2.11 Create Internal R Networks
Utilising igraph [7] the filtered miR-mRNA interactions can be displayed. The colour of the edges represents the level of average correlations between the miRs and mRNAs.
Note. Adjustments to the size of the plot window may be needed to access the key in the bottom left hand side.
Note. Having too many interactions will be very difficult to see in R, so it’s recommend exporting this data into cytoscape, in these cases.
Note. Filtered miR-mRNA interactions must consist of at least 2 miR-mRNA interactions or this plot will not work.
2.12 Output for Pathvisio
MAE3 <- makeMapp(MAE3, filt_df = assay(MAE3, 6), miR_IDs_adj = assay(MAE,
5), dataType = "L")
MAE3 <- makeDynamic(MAE = MAE3, miR_expression = assay(MAE2,
9), mRNA_expression = assay(MAE2, 10), miR_IDs_adj = assay(MAE,
5), dataType = "L")write.table(assay(MAE3, 7), "MAPP.txt", quote = FALSE, row.names = FALSE,
col.names = FALSE, sep = "\t")
write.csv(assay(MAE3, 8), "Dynamics.csv", row.names = TRUE, quote = FALSE)TimiRGeN creates two types of files which can be imported by pathvisio [12]. Firstly, MAPP data which contains miR information. This can be imported into pathvisio via the MAPPbuilder app. This will allow all the miRs to be imported at once. From here some manual adjustments are necessary to properly place the miRs where they are predicted to be effecting the signalling pathway. Next is the Dynamic data which can be imported as an expression data set to visualise the changes that occur to the pathway along the time course.
Please follow the instructions in the ‘inst/Pathvisio_GRN_guide.pdf’ for our method for GRN construction. If any entrezgene/ ensembl IDs are missing, it is recommended to annotate them manually in the MAPP and Dynamics files before importing them into pathvisio. It is recommended to go through pathvisio tutorials [12] for better usage of the network visualisation tool.
2.13 Output to Cytoscape
RCy3::cytoscapePing()
cytoMake(assay(MAE3, 6), titleString = "Lung Fibrosis Pathway",
collectionString = "PathwaysforKidneyFibrosis")The internal R network generated by quickNet can be exported into cytoscape [11] by using functions from RCy3 [5]. TimiRGeN has wrapped around several of these functions to make the process quicker. This will be useful if many miR-mRNA interactions have been found and the internal R graphics cannot cope with this.
Make sure cytoscape version 3.7 or newer is opened, and check if R recognises that cytoscape is open by using using the cytoscapePing() function.
From here further investigation is possible by using cytoscape apps.
3 Separated miR-mRNA Analysis
TimiRGeN also offers the option of analysing the miR and mRNA data separately. Many functions are similar to the combined method, but the "s" mode is used for some functions, rather than the "c" mode.
To go through this we are using a Human Breast Cancer dataset which had Hypoxia induced. This data was published in Cramps et al (2014) [26]. The mRNA data was downloaded from GSE47533 and the miR data was downloaded from GSE47532.
3.1 Load Data
library(org.Hs.eg.db)
miR <- hs_miR
rownames(miR) <- gsub(rownames(miR), pattern = "\\.", replacement = "-")
rownames(miR) <- sub("-$", "*", rownames(miR))For TimiRGeN to work the rownames and colnames must adhere to naming conventions. As mentioned above, the time point should be the first part of each column name and the result type from differential expression analysis e.g. Log2FC, adjPVal or AveEXP should be the second part of the column name. These two parts must be separated by a ., which can be the only . in the column name string.
Furthermore, there should be no _ characters in the column names. Acceptable names: D1.log2fc, H5.ajdpval, TP8.zscore ect. Not acceptable names: D.1.log2fc, H5_adjpval, TP8zscore.
For microRNAs (because naming conventions vary quite a bit), TimiRGeN uses it’s own naming convention which must be adhered to. A - should separate parts of the microRNA name e.g. mmu-miR-193a, mmu-miR-140-5p, hsa-let-7a. The human miR data loaded in this example is from a dataset which uses . instead of -, in the miR names. Depending on your own data, you may have to change the nomenclature to run the data through TimiRGeN, like what was done above.
Some miR data will have a . at the end of the gene names because there can be multiple transcripts with the same name. It is best to keep these extra genes. It is recommended to replace the final . with a * symbol to make this more obvious.
3.2 Differentiate Data
MAE <- addPrefix(MAE = MAE, gene_df = assay(MAE, 1), prefixString = "miR")
MAE <- addPrefix(MAE = MAE, gene_df = assay(MAE, 2), prefixString = "mRNA")addPrefix function will differentiate the miR and mRNA column names. Add a unique prefixString to the miR and mRNA data to do this. This must be used for "s" analysis.
3.3 Find Significantly Differentially Expressed Genes per Timepoint
3.4 Get Significantly Differentially Expressed Gene IDs
3.5 GSEA Method on Separated Data
Here we use ensembl data instead of entrezgene ID. Use the gmtEnsembl function to get wikipathways data which is curated with ensembl IDs.
MAE2 <- gmtEnsembl(MAE = MAE2, path_gene = assay(MAE2, 1), path_data = assay(MAE2,
3), orgDB = org.Hs.eg.db)MAE2 <- enrichWiki(MAE = MAE2, method = "s", ID_list = metadata(MAE)[[4]],
orgDB = org.Hs.eg.db, path_gene = assay(MAE2, 4), path_name = assay(MAE2,
2), ID = "ENSEMBL", universe = assay(MAE2, 4)[[2]])Note. If there are not enough enriched pathways found with the "s" method, GSEA will fail. This occurs often with microRNA data because not many pathways have microRNAs within them. If this occurs, stick to "c" method for miR-mRNA GSEA analysis.
3.6 Temporal Clustering Method on Separated Data
MAE2 <- wikiMatrix(MAE = MAE2, ID_list = metadata(MAE)[[4]],
wp_list = metadata(MAE2)[[2]])
MAE2 <- turnPercent(MAE = MAE2, wikiMatrix = assay(MAE2, 6),
rowInt = 6)Insert miR or mRNA as Data_string for createClusters. This should be the same as the prefixString added before.
MAE2 <- createClusters(MAE = MAE2, method = "s", dataString = "mRNA",
percentMatrix = assay(MAE2, 7), noClusters = 2, variance = 0.99)
clusterCheck(Clusters = metadata(MAE2)[[3]], W = FALSE)
quickFuzz(Mfuzzdata = experiments(MAE2)[[9]], Clusters = metadata(MAE2)[[3]],
W = FALSE)
MAE2 <- returnCluster(MAE2, clusterData = assay(MAE2, 8), whichCluster = 1,
fitCluster = 0.5)From here, once a pathway of interest is found, continue on from section 2.7 to generate miR-mRNA interaction networks.
4 Only mRNA data
If only mRNA or miR data is available, this can also be put through TimiRGeN for time course analysis.
It is easier to run mRNA data by itself, rather than miR data because very few miRs are annotated within signalling pathways.
Note. Conditional data should work here as well.
mRNA <- mm_mRNA
Data <- startObject(miR = NULL, mRNA = mRNA)
Data <- getIdsMrnaMouse(MAE = Data, mRNA = assay(Data, 2), mirror = "useast")
Data <- genesList(Data, method = "c", genetic_data = assay(Data,
2), timeString = "D")
Data <- significantVals(Data, method = "c", geneList = metadata(Data)[[1]],
maxVal = 0.05, stringVal = "adjPVal")
Data <- addIds(MAE = Data, method = "c", filtered_genelist = metadata(Data)[[2]],
miR_IDs = assay(Data, 3), mRNA_IDs = assay(Data, 3))
Data <- eNames(MAE = Data, method = "c", gene_IDs = metadata(Data)[[3]],
ID_Column = 4)Data2 <- enrichWiki(MAE = Data2, method = "c", ID_list = metadata(Data)[[4]],
orgDB = org.Mm.eg.db, path_gene = assay(Data2, 1), path_name = assay(Data2,
2), ID = "ENTREZID", universe = assay(Data2, 1)[[2]])
5 References
- Denise N Slenter et al. “WikiPathways: a multifaceted pathway database bridging metabolomics to other omics research”. In: Nucleic acids research 46.D1 (2018), pp. D661–D667.
- Guangchuang Yu et al. “clusterProfiler: an R package for comparing bi- ological themes among gene clusters”. In: Omics: a journal of integrative biology 16.5 (2012), pp. 284–287.
- Guangchuang Yu et al. “DOSE: an R/Bioconductor package for disease ontology semantic and enrichment analysis”. In: Bioinformatics 31.4 (2014), pp. 608–609.
- Damian Smedley et al. “BioMart–biological queries made easy”. In: BMC genomics 10.1 (2009), p. 22.
- Julia A Gustavsen et al. “RCy3: Network biology using Cytoscape from within R”. In: F1000Research 8 (2019).
- Lokesh Kumar and Matthias E Futschik. “Mfuzz: a software package for soft clustering of microarray data”. In: Bioinformation 2.1 (2007), p. 5.
- Maintainer Gabor Csardi. “Package igraph”. In: Last accessed 3.09 (2013), p. 2013.
- Vikram Agarwal et al. “Predicting effective microRNA target sites in mammalian mRNAs”. In: elife 4 (2015), e05005.
- Yuhao Chen and Xiaowei Wang. “miRDB: an online database for prediction of functional microRNA targets”. In: Nucleic Acids Research.48.D1 (2020), pp D127-D131.
- Hsi-Yuan Huang et al. “miRTarBase 2020: updates to the experimentally validated microRNA–target interaction database.In: Nucleic Acids Research”. 48.D1 (2020), pp D148–D154.
- Michael E Smoot et al. “Cytoscape 2.8: new features for data integration and network visualization”. In: Bioinformatics 27.3 (2010), pp. 431–432.
- Martijn P van Iersel et al. “Presenting and exploring biological pathways with PathVisio”. In: BMC bioinformatics 9.1 (2008), p. 399.
- Alexander R Pico et al. “WikiPathways: pathway editing for the people”. In: PLoS biology 6.7 (2008), e184.
- Hadley Wickham et al. “dplyr: a grammar of data manipulation, 2013”. In: URL https://github.com/hadley/dplyr. version 0.1.[p 1] (2017).
- Gregory R Warnes, Ben Bolker, and Thomas Lumley. “gtools: Various R programming tools”. In: R package version 3.1 (2014).
- Hadley Wickham and Lionel Henry. “Tidyr: Easily tidy data withspread ()andgather ()functions”. In: R package version 0.6 1 (2017).
- Hadley Wickham. “stringr: Make it easier to work with strings”. In: R package version 0.6 2 (2012), pp. 96–7.
- Hadley Wickham. ggplot2: elegant graphics for data analysis. Springer 2016.
- Carlson M (2019). org.Hs.eg.db: Genome wide annotation for Human. R package version 3.8.2.
- Carlson M (2019). org.Mmu.eg.db: Genome wide annotation for Rhesus. R package version 3.8.2.
- Claus Thorn Ekstrom. R Primer. Chapman and Hall/CRC, 2019.
- Ramos M, Schiffer L, Re A, Azhar R, Basunia A, Cabrera CR, Chan T, Chapman P, Davis S, Gomez-Cabrero D, Culhane AC, Haibe-Kains B, Hansen K, Kodali H, Louis MS, Mer AS, Reister M, Morgan M, Carey V, Waldron L (2017). “ Software For The Integration Of Multi-Omics Experiments In Bioconductor.” Cancer Research, 77(21); e39-42.
- Florin L Craciun et al. “RNA sequencing identifies novel translational biomarkers of kidney fibrosis”. In: Journal of the American Society of Nephrology 27.6 (2016), pp. 1702–1713.
- Kathryn L Pellegrini et al. “Application of small RNA sequencing to iden- tify microRNAs in acute kidney injury and fibrosis”. In: Toxicology and applied pharmacology 312 (2016), pp. 42–52.
- Matthias E Futschik and Lokesh Kumar. “Introduction to Mfuzz package and its graphical user interface”. In: (2013).
- Carme Camps et al. “Integrated analysis of microRNA and mRNA expres- sion and association with HIF binding reveals the complexity of microRNA expression regulation under hypoxia”. In: Molecular cancer 13.1 (2014), p. 28.