Quality control and normalization for metabolomics data. Covers QC-based correction, batch effect removal, and data transformation methods. Use when correcting technical variation in metabolomics data before statistical analysis.
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
A direct command skips the review prompt. Inspect the source before running it.
Quality control and normalization for metabolomics data. Covers QC-based correction, batch effect removal, and data transformation methods. Use when correcting technical variation in metabolomics data before statistical analysis.
tool_type
r
primary_tool
MetaboAnalystR
Version Compatibility
Reference examples tested with: xcms 4.0+
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.
Metabolomics Normalization and QC
Load and Inspect Data
Goal: Load the feature table and sample metadata, separating QC and biological samples for downstream processing.
Approach: Read CSV files, partition by sample type, and assess missing value prevalence.
"Normalize my metabolomics data and correct for batch effects" → Apply QC-based signal correction, handle missing values, transform intensities, and assess normalization quality via RSD and PCA.
Goal: Remove injection-order-dependent signal drift using QC sample trends.
Approach: Fit a LOESS curve to QC sample intensities over injection order, then correct all samples by dividing by the predicted drift and rescaling to the QC median.
# QC-based Robust Spline Correction
library(statTarget)
qc_rsc_normalize <-function(data, sample_info){# Fit LOESS to QC samples over injection order# Correct biological samples based on QC trend
injection_order <- sample_info$injection_order
is_qc <- sample_info$sample_type =='QC'
normalized <- data
for(feature in colnames(data)){
qc_values <- data[is_qc, feature]
qc_order <- injection_order[is_qc]# Fit LOESS
fit <- loess(qc_values ~ qc_order, span =0.75)# Predict for all samples
predicted <- predict(fit, injection_order)# Correct: divide by trend, multiply by median
median_val <- median(qc_values, na.rm =TRUE)
normalized[, feature]<- data[, feature]/ predicted * median_val
}return(normalized)}
data_corrected <- qc_rsc_normalize(data, sample_info)
Total Ion Current (TIC) Normalization
Goal: Correct for differences in total signal intensity across samples.
Approach: Divide each sample by its total intensity sum, then rescale to the median total intensity.
Goal: Normalize samples while being robust to large fold changes in individual features.
Approach: Compute a reference spectrum from sample medians, calculate per-sample quotients, and divide each sample by its median quotient.
pqn_normalize <-function(data){# Calculate reference spectrum (median of all samples)
reference <- apply(data,2, median, na.rm =TRUE)# Calculate quotients
quotients <- data / reference
# Normalization factor = median of quotients per sample
factors <- apply(quotients,1, median, na.rm =TRUE)# Normalize
normalized <- data / factors
return(normalized)}
data_pqn <- pqn_normalize(data)
Batch Correction (ComBat)
Goal: Remove systematic technical variation between processing batches while preserving biological effects.
Approach: Apply ComBat empirical Bayes batch correction on log-transformed data, using a design matrix to protect the biological variable of interest.
library(sva)# ComBat for batch correction
batch <- sample_info$batch
mod <- model.matrix(~ sample_info$group)# Keep biological effect# Log transform first
data_log <- log2(data +1)# Apply ComBat
data_combat <- ComBat(dat = t(data_log), batch = batch, mod = mod)
data_combat <- t(data_combat)
Missing Value Handling
Goal: Filter features with excessive missing values and impute remaining gaps for complete-case analysis.
Approach: Remove features missing in more than 20% of samples (optionally per group), then impute via KNN or minimum-value replacement for left-censored data.
# Filter features with too many missing values
filter_missing <-function(data, max_missing =0.2, by_group =TRUE, groups =NULL){if(by_group &&!is.null(groups)){# Keep if present in >80% of samples in at least one group
keep <- sapply(colnames(data),function(f){any(sapply(unique(groups),function(g){
group_data <- data[groups == g, f]
mean(is.na(group_data))<= max_missing
}))})}else{
keep <- colMeans(is.na(data))<= max_missing
}return(data[, keep])}
data_filtered <- filter_missing(data, max_missing =0.2, by_group =TRUE,
groups = sample_info$group)# Impute remaining missing values# KNN imputation
library(impute)
data_imputed <- impute.knn(as.matrix(data_filtered), k =5)$data
# Or minimum value imputation (for left-censored data)
min_impute <-function(data){
data_imp <- data
for(col in colnames(data)){
min_val <-min(data[, col], na.rm =TRUE)/2
data_imp[is.na(data_imp[, col]), col]<- min_val
}return(data_imp)}
Data Transformation
Goal: Transform and scale feature intensities to approximate normality and equalize feature variance.
Approach: Apply log2 transformation followed by Pareto scaling (divide by sqrt of SD) or auto-scaling (z-score).
