LipidSig: a web-based tool for lipidomic data analysis

In Differential Expression Page, significant lipid species or lipid characteristics can be explored through two main customised analysis, by ‘Lipid species’ or by ‘Lipid characteristics’ , with user-uploaded data. Subsequently, further analysis and visualisation methods, including dimensionality reduction, hierarchical clustering, characteristics analysis, and enrichment, can be implemented based on the results of differential expressed analysis by utilising user-defined methods and characteristics.

Data Source

Demo dataset

Adipose tissue ATGL modifies the cardiac lipidome in pressure-overload-induced left Ventricular failure (PLoS Genet. 2018)

Lipid dataset can be uploaded by users or using example datasets. This information, namely Lipid expression data , Group information , and Lipid characteristics , needs to be uploaded in CSV format.

Human plasma lipidome from 10 healthy controls and 13 patients with systolic heart failure (HFrEF) were analyzed by MS-based shotgun lipidomics. The data revealed dysregulation of individual lipid classes and lipid species in the presence of HFrEF.

Group information

Lipid characteristics

: Successfully uploaded. : Error happaned. Please check your dataset. : Warning message.

Lipid expression data

Processed data
Uploaded data

Group information

Lipid characteristics

Result

Lipid species analysis
Lipid characteristics analysis

Differential expression

In lipid species analysis section, differentially expressed analysis is performed to find significant lipid species. In short, samples will be divided into two groups (independent) based on the Group Information of input. Two statistical methods, t-test and Wilcoxon test (Wilcoxon rank-sum test), are provided and p-value will be adjusted by Benjamini-Hochberg procedure. The condition and cut-offs for significant lipid species are also users selected.

Method:

Multiple testing correction:

Benjamini & Hochberg

Identify significant lipids:

p-value:

Fold change (FC):

Dimensionality reduction

Dimensionality reduction is common when dealing with large numbers of observations and/or large numbers of variables in lipids analysis. It transforms data from a high-dimensional space into a low-dimensional space so that to retain vital properties of the original data and close to its intrinsic dimension.

Dimensionality reduction method:

Scaling:

Centering:

NOTE: Scaling and Centering are on by default. Centering is strongly recommended for pre-processing steps.

Scaling:

PCA:

Perplexity:

Number of iterations:

Scaling:

Number of neighbors:

Distance metric:

Clustering method:

Number of groups:

Distance metrics:

Number of groups:

Distance measure:

Clustering method:

Epsilon:

minPts:

Hierarchical clustering

Lipid species that derived from two groups will be clustered and visualised on heatmap using hierarchical clustering. Through heatmap, users may discover the difference between the two groups by observing the distribution of lipid species. This analysis provides an overview of lipid species differences between the control group and the experimental group.

Select feature for clustering:

Side color by lipid characteristics:

Distance measure:

Clustering method:

Download matrix

Characteristics association

In this part, we categorize significant lipid species based on different lipid characteristics and visualise the difference between control and experimental groups by applying log2 Fold Change.

Select the lipid characteristics:

Enrichment

Enrichment plot assists users to determine whether significant lipid species are enriched in the categories of the selected characteristics. The enrichment plots and a summary table are further classified into up/down/all groups by log2 fold change of significant lipid species. Each group (value) of the selected characteristics will have a value of significant count and p-value within a summary table.

Method:

Fisher's exact test

Select the lipid characteristics:

p-value:

KEGG pathway analysis

Differential expression

In Lipid Characteristics Analysis section, lipid species are categorised and summarised into new lipid expression table according to two selected lipid characteristic, then conducted differential expressed analysis. Samples will be divided into two groups based on the Group Information of input data. Two-way ANOVA is applied with t-test as post hoc tests. This Differentially Expressed Analysis section separates into 2 sections, analysing based on first ‘Characteristics’ and adding ‘Subgroup of characteristics’ to the analysis. The first section is analysed based on the first selected ‘characteristics’ . The second section is the subgroup analysis of the first section.

Characteristics:

Method:

Two-way ANOVA

Post hoc tests:

t-test

Subgroup of characteristics:

NOTE: two selected characteristics should be both continuous data or one categorical data with one continuous data.

p-value:

Fold change (FC):

Raw
Sqrt scale

Subgroup of characteristics:

This is the ‘Subgroup of characteristics’ section. In short, lipid species will be further split by the characteristic that user-chosen in the second pull-down menu then undergo the first section analysis. Two-way ANOVA is also applied with t-test as post hoc tests, and the cut-offs of differentially expressed lipids are inputted by users.

Select a category:

Raw
Sqrt scale

Dimensionality reduction

Dimensionality reduction in this section assists users to tackle with large numbers of variables in lipids analysis. The high-dimensional space is transformed into a low-dimensional space. Hence, the crucial properties of the lipid data are revealed and still close to its intrinsic characteristics. Here, we provide four types of dimensionality reduction approaches, PCA, PLS-DA, t-SNE, UMAP, and four clustering methods, K-means, partitioning around medoids (PAM), Hierarchical clustering, and DBSCAN.

Dimensionality reduction method:

Scaling:

Centering:

NOTE: Scaling and Centering are on by default. Centering is strongly recommended for pre-processing steps.

Scaling:

PCA:

Perplexity:

Number of iterations:

Scaling:

Number of neighbors:

Distance metric:

Clustering method:

Number of groups:

Distance metrics:

Number of groups:

Distance measure:

Clustering method:

Epsilon:

minPts:

Hierarchical clustering

New lipid expression table summed up from species will be clustered and shown on the heatmap using hierarchical clustering. Through heatmap, users may discover the difference between the two groups by observing the distribution of lipid characteristic expression. This analysis provides an overview of lipid characteristic expression differences between the control group and the experimental group. Four distance measures can be chosen, Person, Spearman, or Kendall, and eight clustering methods can be selected by pulling down the menu.

Select feature for clustering:

Distance measure:

Clustering method:

Download matrix