Dexma

DExMA

Workflows

Standard Workflow

Perform a complete gene expression meta-analysis workflow including data preparation, annotation harmonization, log transformation, heterogeneity testing, optional imputation, meta-analysis, and heatmap visualization.

library(DExMA)

# Create meta-analysis object
phenoGroups <- c("condition", "condition", "state", "state")
phenoCases <- list(Study1 = "Diseased", Study2 = c("Diseased", "ill"),
                   Study3 = "Diseased", Study4 = "ill")
phenoControls <- list(Study1 = "Healthy", Study2 = c("Healthy", "control"),
                      Study3 = "Healthy", Study4 = "control")

newObjectMA <- createObjectMA(listEX = listMatrixEX,
                              listPheno = listPhenodatas,
                              namePheno = phenoGroups,
                              expGroups = phenoCases, 
                              refGroups = phenoControls)

# Harmonize gene IDs
newObjectMA <- allSameID(newObjectMA, finalID = "GeneSymbol", organism = "Homo sapiens")

# Log-transform data
newObjectMA <- dataLog(newObjectMA)

# Test for heterogeneity
heterogeneityTest(newObjectMA)

# Perform meta-analysis
resultsMA <- metaAnalysisDE(newObjectMA, typeMethod = "REM", missAllow = 0.3, proportionData = 0.50)

# Visualize results
makeHeatmap(objectMA = newObjectMA, resMA = resultsMA, scaling = "zscor", 
            regulation = "all", numSig = 40, fdrSig = 0.05, logFCSig = 1.5, show_rownames = TRUE)

Inputs are lists of expression matrices and phenodata; output is a meta-analysis results table and a heatmap of significant genes.

Batch Effect Correction

Detect and remove batch effects from individual datasets and calculate individual study effect sizes or p-values.

library(DExMA)

# Remove batch effects
corrected_study2 <- batchRemove(listMatrixEX$Study2, listPhenodatas$Study2, 
                                formula = ~gender+race, 
                                mainCov = "race", nameGroup = "condition")

# Calculate individual study effect sizes
effects <- calculateES(newObjectMA)

# Calculate individual study p-values
pvalues <- pvalueIndAnalysis(newObjectMA)

Inputs are expression matrices and phenodata; outputs are batch-corrected expression matrices, individual study effect sizes, and individual p-values.

Download Geo Data

Download public microarray datasets from NCBI GEO for meta-analysis.

library(DExMA)

# Download GEO datasets
GEOobjects <- c("GSE4588", "GSE10325")
# dataGEO <- downloadGEOData(GEOobjects)

Input is a vector of GEO accession IDs; output is a list of downloaded GEO datasets.

When to Use

• Performing gene expression meta-analysis across multiple microarray or RNA-Seq datasets using metaAnalysisDE.
• Harmonizing heterogeneous gene identifiers across different studies to a common standard (e.g., Official Gene Symbol) using allSameID.
• Imputing missing gene expression values across studies using the KNN method via missGenesImput.
• Removing batch effects from individual expression matrices using batchRemove.

When NOT to Use

• For single-dataset differential expression analysis, use limma, DESeq2, or edgeR directly because DExMA is designed for multi-study meta-analysis.
• For raw RNA-Seq read alignment or quantification, use packages like Rsubread because DExMA requires pre-normalized expression matrices.

Data Requirements

• Input Format: A list of nested lists (an objectMA created via createObjectMA or elementObjectMA).
• Structure: Each study must contain an expression matrix (genes in rows, samples in columns) and a binary vector (0 for reference/control, 1 for experimental/cases).
• Normalization: Expression data should be normalized and log-transformed (can use dataLog for log-transformation).

Key Parameters

• listEX: A list of expression dataframes, matrices, or ExpressionSet objects.
• listPheno: A list of phenodata dataframes.
• finalID: The target gene ID type (e.g., "GeneSymbol") in allSameID.
• organism: The organism name (e.g., "Homo sapiens") in allSameID.
• typeMethod: The meta-analysis method (e.g., "REM" for random effects model, "maxP" for p-value combination) in metaAnalysisDE.
• missAllow (0.3): The maximum proportion of missing values allowed for a gene.
• proportionData (0.50): The minimum proportion of datasets in which a gene must be present.

Best Practices

• Harmonize all dataset gene IDs to a common annotation (e.g., Official Gene Symbol) using allSameID before performing meta-analysis.
• Apply dataLog to ensure all datasets are in log scale.
• Assess study heterogeneity using heterogeneityTest to decide between fixed-effects and random-effects models.
• Use missGenesImput to impute missing genes if you want to avoid discarding genes that are unmeasured in a subset of studies.

Common Pitfalls

• Mixing log-transformed and non-log-transformed datasets: Use dataLog to automatically detect and log-transform non-log datasets.
• Discarding too many genes due to missingness: Adjust missAllow and proportionData in metaAnalysisDE, or use missGenesImput to impute missing values.
• Inconsistent group definitions across studies: Carefully construct phenoCases and phenoControls lists when calling createObjectMA.

Alternatives

• limma for single-study differential expression and batch correction.
• swamp for principal component analysis and batch effect identification (e.g., prince).
• Biobase for managing ExpressionSet objects.

Citations

• Villatoro-García, J. A., & Carmona-Sáez, P. (2026). Differential Expression Meta-Analysis with DExMA package.

References

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