Mirtarrnaseq

mirTarRnaSeq

Workflows

Mirna Mrna Correlation Multi Timepoint

Identify significant miRNA-mRNA correlations across three or more time points using fold change data and background distribution comparisons.

library(mirTarRnaSeq)

# 1. Parse fold changes and filter mRNAs/miRNAs
mrna <- one2OneRnaMiRNA(mrna_files, pthreshold = 0.05)$foldchanges
mirna <- one2OneRnaMiRNA(mirna_files)$foldchanges

# 2. Calculate Pearson correlation
corr_0 <- corMirnaRna(mrna, mirna, method = "pearson")

# 3. Generate background correlation distribution
outs <- sampCorRnaMirna(mrna, mirna, method = "pearson", Shrounds = 100, Srounds = 1000)

# 4. Plot correlation density
mirRnaDensityCor(corr_0, outs)

# 5. Identify significant correlations
sig_corrs <- threshSig(corr_0, outs, pvalue = 0.05)

# 6. Retrieve species-specific miRanda predictions
miRanda <- getInputSpecies("Mouse", threshold = 150)

# 7. Intersect and visualize results
results <- miRandaIntersect(sig_corrs, outs, mrna, mirna, miRanda)
p <- mirRnaHeatmap(results$corr, upper_bound = -0.99)

Input: Multi-timepoint mRNA and miRNA fold change files. Output: Heatmap and list of significant miRNA-mRNA correlations intersected with miRanda predictions.

Mirna Mrna Relationship Two Timepoint

Identify significant miRNA-mRNA fold change differences between two time points compared to a background distribution.

library(mirTarRnaSeq)

# 1. Parse fold changes and filter mRNAs
mrna <- one2OneRnaMiRNA(mrna_files, pthreshold = 0.05)$foldchanges
mirna <- one2OneRnaMiRNA(mirna_files)$foldchanges

# 2. Estimate miRNA-mRNA fold change differences
inter0 <- twoTimePoint(mrna, mirna)

# 3. Generate background distribution of fold change differences
outs <- twoTimePointSamp(mrna, mirna, Shrounds = 10)

# 4. Retrieve species-specific miRanda target predictions
miRanda <- getInputSpecies("Mouse", threshold = 140)

# 5. Identify relationships below a p-value threshold
sig_InterR <- threshSigInter(inter0, outs)

# 6. Intersect significant relationships with miRanda predictions
results <- mirandaIntersectInter(sig_InterR, outs, mrna, mirna, miRanda)

# 7. Format final results and draw plots
final_results <- finInterResult(results)
drawInterPlots(mrna, mirna, final_results)
mirRnaHeatmapDiff(results$corrs, upper_bound = 9.9)

Input: Two-timepoint mRNA and miRNA fold change files. Output: Fold change plots and heatmap of significant miRNA-mRNA differences.

Standard Workflow

Identify miRNA-mRNA target interactions in a cohort using various GLM regression models integrated with miRanda predictions.

library(mirTarRnaSeq)

# 1. Retrieve species-specific miRanda target predictions
miRanda <- getInputSpecies("Epstein_Barr", threshold = 140)

# 2. Filter mRNA data to keep only predicted targets
DiffExpmRNASub <- miRanComp(DiffExp, miRanda)

# 3. Combine mRNA and miRNA data
miRNA_select <- c("ebv-mir-BART9-5p")
Combine <- combiner(DiffExp, miRNAExp, miRNA_select)
geneVariant <- geneVari(Combine, miRNA_select)

# 4. Run univariate GLM models (e.g., Poisson)
j <- runModel(`LMP-1` ~ `ebv-mir-BART9-5p`, Combine, model = glm_poisson(), scale = 100)
print(modelTermPvalues(j))

# 5. Run models across all combinations with a specific family
blaGaus <- runModels(Combine, geneVariant, miRNA_select, family = glm_gaussian(), scale = 100)

# 6. Run all-to-all combinations using multi-model GLM
All_miRNAs_run <- runAllMirnaModels(
  mirnas = rownames(miRNAExp)[1:5],
  DiffExpmRNA = DiffExpmRNASub,
  DiffExpmiRNA = miRNAExp,
  miranda_data = miRanda,
  prob = 0.75,
  cutoff = 0.05,
  fdr_cutoff = 0.1,
  method = "fdr",
  family = glm_multi(),
  scale = 2,
  mode = "multi"
)

Input: mRNA and miRNA expression matrices. Output: Fitted GLM models and identified significant miRNA-mRNA target interactions.

When to Use

• To perform regression analysis (Gaussian, Poisson, Negative Binomial, Zero-Inflated) between miRNA and mRNA expression across sample cohorts using runModels().
• To identify significant miRNA-mRNA correlations across 3 or more time points using corMirnaRna() and sampCorRnaMirna().
• To compare miRNA-mRNA fold change differences between two time points using twoTimePoint() and twoTimePointSamp().
• To filter and intersect statistical results with species-specific miRanda target predictions using getInputSpecies() and miRanComp().

When NOT to Use

• For predicting physical miRNA-mRNA binding sites from sequence data alone; use standalone miRanda or TargetScan because mirTarRnaSeq requires expression data to perform statistical association.
• For differential expression analysis of RNA-seq data; use DESeq2 or edgeR because mirTarRnaSeq expects pre-computed expression matrices or fold changes.

Data Requirements

• mRNA and miRNA expression matrices (TPM, RPKM, or normalized counts) with samples as columns and gene/miRNA names as rows.
• For time-point analyses, pre-computed fold change tables from differential expression analysis.
• Species-specific miRanda target prediction files (supported species include "Human", "Mouse", "Epstein_Barr", etc.).

Key Parameters

• threshold (140): Score threshold for filtering miRanda predictions in getInputSpecies().
• scale (100): Scaling factor applied to expression values during GLM modeling.
• family (glm_gaussian()): Statistical distribution family for GLM modeling (e.g., glm_poisson(), glm_nb(), glm_zeroinfl(), glm_multi()).
• mode ("multi"): Mode for running multi-model GLMs ("multi" or "inter").
• pthreshold (0.05): P-value threshold for filtering mRNAs in one2OneRnaMiRNA().
• Shrounds (100): Number of shuffling rounds for generating background distributions.

Best Practices

• Normalize both mRNA and miRNA expression matrices (e.g., using TPM or z-score normalization via tzTrans()) before running regression models.
• Compare Akaike Information Criterion (AIC) values across different GLM families (Gaussian, Poisson, Negative Binomial) to select the best-fitting model for your data.
• Filter the mRNA expression matrix using miRanComp() prior to running all-to-all regressions to reduce computational time.

Common Pitfalls

• High computational time: Running all-to-all regressions on unfiltered genome-wide mRNA matrices is extremely slow. Fix by filtering mRNAs to keep only predicted targets of selected miRNAs using miRanComp().
• Incompatible sample names: If sample names (column names) do not match exactly between mRNA and miRNA expression matrices, combiner() will fail. Fix by aligning column names before analysis.

Alternatives

• multiMiR for retrieving validated and predicted miRNA-target interactions from multiple databases.
• SPONGE for sparse partial correlation and miRNA-mRNA competitive endogenous RNA (ceRNA) networks.
• DESeq2 / edgeR for standard differential expression analysis of mRNA and miRNA.

Citations

• Movassagh et al. 2019, Scientific Reports (for EBV stomach cancer dataset)

References

• Homepage: https://bioconductor.org/packages/mirtarrnaseq
• Vignette: https://bioconductor.org/packages/release/bioc/vignettes/mirtarrnaseq/inst/doc/mirTarRnaSeq.html