knitr::opts_chunk$set(echo = TRUE, message = FALSE, warning = FALSE, fig.width = 8)


lipidr implements a series of functions to facilitate inspection, analysis and visualization of targeted lipidomics datasets. lipidr takes exported Skyline CSV as input, allowing for multiple methods to be analyzed together.


lipidr represents Skyline files as SummarizedExperiment objects, which can easily be integrated with a wide variety of Bioconductor packages. Sample annotations, such as sample group or other clinical information can be loaded.

Quality control & visualization

lipidr generates various plots, such as PCA score plots and box plots, for quality control of samples and measured lipids. Normalization methods with and without internal standards are also supported.

Differential Analysis

Differential analysis can be performed using any of the loaded clinical variables, which can be readily visualized as volcano plots. A novel lipid set enrichment analysis (LSEA) is implemented to detect preferential enrichment of certain lipid classes, total chain lengths or unsaturation patterns. Plots for the visualization of enrichment results are also implemented.

This vignette provides a step by step guide for downstream analysis of targeted lipidomics data, exported from Skyline.


From Bioconductor

In R console, type:

if (!requireNamespace("BiocManager", quietly=TRUE))

From GitHub using devtools

In R console, type:


Analysis Workflow

Example Study

In this workflow, we will use serum lipidomics data from mice fed a normal or high-fat diet. Mice were fed a normal or high-fat diet (Diet column) and had access to normal drinking water or drinking water containing the bile acid deoxycholic acid (BileAcid column). Lipid peaks were integrated using Skyline and exported as CSV files.

Data Import & manipulation

Exporting files from Skyline

Integrated peaks should be exported from each Skyline file through File => Export => Report. Selecting Transition Results ensures that necessary information is exported from Skyline. Otherwise, you should ensure that peak Area or Height or a similar measure is exported. Regardless of the measure you choose for intensity, you can use lipidr workflow.

Replicates should either be exported, or the Pivot Replicate Name option must be used.

Reading files into R

lipidr can read multiple CSV files from different analysis methods together. Using our example dataset, three Skyline CSV files are used as input to read.skyline.

datadir = system.file("extdata", package="lipidr")
filelist = list.files(datadir, "data.csv", full.names = TRUE) # all csv files

d = read_skyline(filelist)

Datasets are represented in R as SummarizedExperiments to facilitate integration other Bioconductor packages.

Adding sample annotation

Sample annotation can be prepared in Excel and saved as CSV. The table should have at least two columns, first indicating sample names and other columns indicating clinical variables.

clinical_file = system.file("extdata", "clin.csv", package="lipidr")

d = add_sample_annotation(d, clinical_file)

Data subsetting

It is helpful to imagine LipidomicsExperiment object as a table with lipid molecules as rows and samples as columns. We can subset this table by selecting specific rows and columns. The general syntax is d[rows, cols].

In the example below we select the first 10 transitions and 10 samples. We can check the rowData and colData.

d_subset = d[1:10, 1:10]

We can also apply conditional selections (indexing). For example, we can select all quality control samples.

d_qc = d[, d$group == "QC"]

Note that we leave rows index empty (d[,cols]) to select all lipids. We can also subset based on lipid annotations, selecting a specific class for example.

pc_lipids = rowData(d)$Class %in% c("PC", "PCO", "PCP")
d_pc = d[pc_lipids,]

For demonstration purposes, we select only 3 lipids classes, Ceramides (Cer), PhosphatidylCholines (PC) and LysoPhosphatidylCholines (LPC). We also BileAcid treated group from our dataset.

lipid_classes = rowData(d)$Class %in% c("Cer", "PC", "LPC")
groups = d$BileAcid != "DCA"
d = d[lipid_classes, groups]

#QC sample subset
d_qc = d[, d$group == "QC"]

Raw Data Quality Check

To ensure data quality, we can look at total lipid intensity as bar chart or distribution of samples as a boxplot.

plot_samples(d, type = "tic", log = TRUE)

We can also look at intensity and retention time distributions for each lipid molecule using plot_molecules(type = boxplot). It is recommended to assess the variation across quality control samples.

plot_molecules(d_qc, "sd", measure = "Retention Time", log = FALSE)
plot_molecules(d_qc, "cv", measure = "Area")

Or intensity distribution within different lipid classes.

plot_lipidclass(d, "boxplot")

Interactive plots

All lipidr plots can be displayed interactive mode if plotly package is installed. Plot interactivity is disabled by default. To enable interactivity, simple call use_interactive_graphics(). You can turn interactivity back off by use_interactive_graphics(interactive=FALSE).

Summarizing transitions

This step is important when more than one transition is measured per lipid molecule. Multiple transitions are summarized into a single value by either taking the average intensity or the one with highest intensity.

d_summarized = summarize_transitions(d, method = "average")


Probabilistic Quotient Normalization (PQN)

The PQN method determines a dilution factor for each sample by comparing the distribution of quotients between samples and a reference spectrum, followed by sample normalization using this dilution factor.

d_normalized = normalize_pqn(d_summarized, measure = "Area", exclude = "blank", log = TRUE)
plot_samples(d_normalized, "boxplot")

By specifying exclude = "blank", blank runs are automatically detected and excluded from the normalization process.

Internal standard normalization

Internal standard normalization corrects lipid class-specific variations between samples. Lipid classes are normalized using corresponding internal standard(s) of the same lipid class. If no corresponding internal standard is found the average of all measured internal standards is used instead.

d_normalized_istd = normalize_istd(d_summarized, measure = "Area", exclude = "blank", log = TRUE)

Multivariate analysis

You can investigate sample variation using either PCA or PCoA (classical MDS).

mvaresults = mva(d_normalized, measure="Area", method="PCA")
plot_mva(mvaresults, color_by="group", components = c(1,2))

Plotting other components is possible by specifying components argument. For example components = c(2,3) plots second and third components.

Supervised multivariate analysis

Supervised multivariate analyses, such as OPLS and OPLS-DA can be performed to determine which lipids are associated with a group (y-variable) of interest. In this example we use "Diet" as grouping, and display the results in a scores plot.

mvaresults = mva(d_normalized, method = "OPLS-DA", group_col = "Diet", groups=c("HighFat", "Normal"))
plot_mva(mvaresults, color_by="group")

We can also plot the loadings and display important lipids contributing to the separation between different (Diet) groups.

plot_mva_loadings(mvaresults, color_by="Class", top.n=10)

Alternatively, we can extract top N lipids along with their annotations.

top_lipids(mvaresults, top.n=10)

Differential analysis

This step of the workflow requires the limma package to be installed.

Normalized and log transformed data should be used.

de_results = de_analysis(
  HighFat_water - NormalDiet_water,

We can visualize differential analysis results using volcano plots. In this instance we turn off labeling, due to the large number of significantly altered lipids between conditions.

plot_results_volcano(de_results, show.labels = FALSE)

Complex experimental designs

For more complex experimental designs, where one might deal with factor adjustment or interactions, it is recommended to use the de_design function. Users can either provide a design matrix, or a formula to create one.

# Using formula
  design = ~ group, 

# Using design matrix
design = model.matrix(~ group, data=colData(d_normalized))
  design = design,

For more details on creating design matrices for different experimental designs, refer to (Limma User Guide)[] and (edgeR tutorial)[].

Enrichment analysis

lipidr automatically generates sets of lipids based on lipid class, total chain length and unsaturation. Measured lipids are then ranked by their fold change, or p-value using results from differential analysis.

enrich_results = lsea(de_results, = "logFC")

Visualization of enrichment analysis results. The enriched lipid classes and chain unsaturations are highlighted.

plot_enrichment(de_results, significant_lipidsets(enrich_results), annotation="class")
plot_enrichment(de_results, significant_lipidsets(enrich_results), annotation="unsat")

lipidr can also plot fold changes of lipids per class showing total chain lengths and unsaturations. Number on the plot indicate multiple lipids measured with the same chain properties.


Session information


ahmohamed/lipidr documentation built on July 7, 2023, 2:22 a.m.