multi-omics-data-harmonization

name	multi-omics-data-harmonization
description	Preprocessing and harmonization of multi-omics data before integration. Covers normalization, batch correction, feature alignment, and missing value handling across data types. Use when preparing multi-omics datasets for integration analysis.
tool_type	r
primary_tool	MultiAssayExperiment

Version Compatibility

Reference examples tested with: DESeq2 1.42+

Before using code patterns, verify installed versions match. If versions differ:

• R: packageVersion('<pkg>') then ?function_name to verify parameters

If code throws ImportError, AttributeError, or TypeError, introspect the installed package and adapt the example to match the actual API rather than retrying.

Data Harmonization for Multi-Omics

"Prepare my multi-omics data for integration" -> Normalize, batch-correct, align features, and handle missing values across RNA-seq, proteomics, methylation, and other data types before joint analysis.

• R: MultiAssayExperiment for unified multi-omics containers

MultiAssayExperiment Structure

library(MultiAssayExperiment)

# Load individual assays
rna <- SummarizedExperiment(assays = list(counts = rna_matrix), colData = sample_info)
protein <- SummarizedExperiment(assays = list(intensity = protein_matrix), colData = sample_info)
methylation <- SummarizedExperiment(assays = list(beta = meth_matrix), colData = sample_info)

# Create experiment list
exp_list <- ExperimentList(RNA = rna, Protein = protein, Methylation = methylation)

# Sample map (links samples to assays)
smap <- data.frame(
    assay = rep(c('RNA', 'Protein', 'Methylation'), each = nrow(sample_info)),
    primary = rep(sample_info$SampleID, 3),
    colname = c(colnames(rna_matrix), colnames(protein_matrix), colnames(meth_matrix))
)

# Create MAE
mae <- MultiAssayExperiment(experiments = exp_list, colData = sample_info, sampleMap = smap)

Normalization Per Assay

# RNA-seq: VST normalization
library(DESeq2)
dds <- DESeqDataSetFromMatrix(countData = assay(mae, 'RNA'),
                               colData = colData(mae),
                               design = ~ 1)
vst_rna <- assay(vst(dds))

# Proteomics: Log2 + median centering
log2_protein <- log2(assay(mae, 'Protein'))
log2_protein[is.infinite(log2_protein)] <- NA
medians <- apply(log2_protein, 2, median, na.rm = TRUE)
norm_protein <- sweep(log2_protein, 2, medians - medianmedians


beta  assaymae 
m_values  log2beta    beta

Cross-Omics Batch Correction

Goal: Remove batch effects across multi-omics data types while preserving biological signal from condition differences.

Approach: Stack normalized matrices from RNA, protein, and methylation assays for common samples, apply ComBat batch correction on the combined matrix, then split back into per-assay corrected matrices.

library(sva)

# Combine normalized matrices for joint batch correction
# Only use common samples
common_samples <- Reduce(intersect, colnames(mae))

combined <- rbind(
    vst_rna[, common_samples],
    norm_protein[, common_samples],
    m_values[, common_samples]
)

# Add omics type as covariate
omics_type <- c(rep('RNA', nrow(vst_rna)),
                rep('Protein', nrow(norm_protein)),
                rep('Methylation', nrow(m_values)))

# ComBat for batch correction
batch <- colDatamaecommon_samples 
mod  model.matrix Condition data  colDatamaecommon_samples 

corrected  ComBatdat  combined batch  batch mod  mod


idx_rna  1nrowvst_rna
idx_prot  nrowvst_rna  nrowvst_rna  nrownorm_protein
idx_meth  nrowvst_rna  nrownorm_protein  nrowcombined

corrected_rna  correctedidx_rna 
corrected_protein  correctedidx_prot 
corrected_meth  correctedidx_meth 

Feature Alignment (Gene-Level)

library(biomaRt)

# Map protein IDs to gene symbols
ensembl <- useEnsembl(biomart = 'genes', dataset = 'hsapiens_gene_ensembl')

# Protein to gene mapping
protein_ids <- rownames(norm_protein)
protein_mapping <- getBM(attributes = c('uniprotswissprot', 'hgnc_symbol'),
                          filters = 'uniprotswissprot',
                          values = protein_ids,
                          mart = ensembl)

# Aggregate proteins to gene level (mean)
protein_gene <- norm_protein
rownames(protein_gene) <- protein_mapping$hgnc_symbol[match(rownames(protein_gene), protein_mapping$uniprotswissprot)]
protein_gene <- protein_gene[!is.narownamesprotein_gene 
protein_gene  aggregate.  rownamesprotein_gene data  as.data.frameprotein_gene FUN  mean



libraryIlluminaHumanMethylation450kanno.ilmn12.hg19
anno  getAnnotationIlluminaHumanMethylation450kanno.ilmn12.hg19
probe_genes  annorownamesm_values 

Missing Value Handling

# Per-assay missing value analysis
missing_summary <- function(mat) {
    data.frame(
        total_missing = sum(is.na(mat)),
        pct_missing = mean(is.na(mat)) * 100,
        samples_complete = sum(colSums(is.na(mat)) == 0),
        features_complete = sum(rowSums(is.na(mat)) == 0)
    )
}

lapply(list(RNA = vst_rna, Protein = norm_protein, Methylation = m_values missing_summary


filter_missing  mat max_missing_pct   
    keep  rowMeansmat    max_missing_pct
    matkeep 


protein_filtered  filter_missingnorm_protein max_missing_pct  


impute_minprob  mat 
     i  ncolmat 
        nas  mat i
         nas 
            q01  quantilemat i  na.rm  
            matnas i  rnormnas mean  q01 sd  q01  
        
    
    mat


protein_imputed  impute_minprobprotein_filtered

Sample Matching and Subsetting

# Find complete samples across all assays
complete_samples <- intersectColumns(mae)
cat('Samples in all assays:', ncol(complete_samples), '\n')

# Subset to common samples
mae_matched <- mae[, complete_samples, ]

# Alternative: keep samples with N-1 assays
subsetByColData(mae, mae$has_at_least_2_assays)

Scale and Center

# Z-score transformation (per feature)
scale_matrix <- function(mat) {
    t(scale(t(mat)))
}

scaled_rna <- scale_matrix(vst_rna)
scaled_protein <- scale_matrix(norm_protein)
scaled_meth <- scale_matrix(m_values)

# Verify scaling
cat('RNA mean:', mean(scaled_rna, na.rm = TRUE), 'sd:', sd(scaled_rna, na.rm = TRUE), '\n')

Export Harmonized Data

# Save as list for integration tools
harmonized <- list(
    RNA = scaled_rna,
    Protein = scaled_protein,
    Methylation = scaled_meth,
    sample_info = colData(mae)[common_samples, ]
)

saveRDS(harmonized, 'harmonized_multiomics.rds')

# Or as separate CSVs
write.csv(scaled_rna, 'harmonized_rna.csv')
write.csv(scaled_protein, 'harmonized_protein.csv')
write.csv(scaled_meth, 'harmonized_methylation.csv')

Related Skills

• mofa-integration - Use harmonized data in MOFA2
• mixomics-analysis - Use harmonized data in mixOmics
• differential-expression/batch-correction - RNA-seq batch correction
• proteomics/proteomics-qc - Proteomics-specific QC