This page works best on a computer screen
Analyses were done using a metabolomics and a lipidomics dataset created by Metabolon Inc., US. Select the dataset of interest below.

Metabolomics dataset

The following figures are the results of the WGCNA (Weighted Gene Co-expression Network Analysis) run using metabolomics data from patients hospitalised with influenza or SARS-CoV-2. The metabolites are measured as relative abundances which were batch-normalised, minimum value imputed, and log2-transformed to obtain normal distributions with medians of 0. During data pre-processing, non-varied metabolites were removed.

Modules

Groups of metabolites are found using the WGCNA package (link). First, correlations between all metabolites were calculated using the bi-weight mid-correlation measure which is median-based and less sensitive to outliers. Next, the topological overlap of the correlations was found. This measure takes into account both the molecule-to-molecule correlations, but also the two molecules’ shared correlations, which is a way to simulate biological pathway structures. Hierarchical clustering of the topological overlap measures was done to find groups of metabolites with similar abundance patterns across the patient cohort. Clustering of the metabolomics data resulted in 14 modules whereof one, the grey module, contains the metabolites that did not fit in any of the remaining clusters. All modules are represented by a colour and can be seen in the dendrogram below where each branch repressents a metabolite. Keep in mind that this is a 2D schematic of a multidimensional figure and each “hinge” in the tree can be flipped.

Correlations with clinical variables

Spearman’s Rank Correlations were calculated for the weighted average abundances (first principal component) of all metabolites in each module association clinical variables. Red indicate a positive correlation and blue a negative one. NB: not all metabolites in a module necessarily show the same correlation. The asterisks indicate false discovery rate (FDR) adjusted p-values: * = q<0.05, ** = q<0.01, *** = q<0.0001. The modules are represented by the colour column to the right and the dendrogram indicate distances between the clusters.















Super-pathway content of modules

Each metabolite was linked to a metabolic super-pathway and sub-pathway based on the biological function of the molecule (this was done by Metabolon Inc., US). NB: Each molecule is only repressented by one pathway although they may biologically be active in multiple.
The figure below shows the super-pathway content in each modules of metabolites as a percentage, e.g., ~20% of the green cluster consists of amino acids. The legend features the total amount of metabolites measured in our samples that have been annotated to each super-pathway category.

Super- and sub-pathway content of modules

In the tabs below are figures showing the super- and sub-pathway content of each module created with WGCNA. Hovering over the figure will show which metabolites are in each category.
NB: Some modules (e.g. the grey) contain too many metabolites to produce a nice figure. The full content of all modules can be found in the table at the end.

green

salmon

blue

purple

red

turquoise

black

brown

tan

greenyellow

magenta

pink

yellow

grey

Table of medians for progressors and non-progressors

The table includes all metabolites from the two modules, red and turquoise, which were associated with disease progression after adjustment for confounders and with a FDR p-value cutoff of 0.05. The metabolites were included in the WGCNA as batch normalised and log2-transformed abundances, meaning each metabolite has a median of 0. The median for each metabolite is presented for progressors and non-progressors. Positive values means a metabolite abundance for this group of samples above the median for the whole cohort, and negative means below the overall median.
NB: It is possible for metabolites placed in the same cluster to have opposite associations with the same clinical variable due to the unsupervised clustering approach.

Table of metabolites in modules

All measured metabolites can be found in the table below along with information about the molecule and which cluster it was placed withing. You can type in search terms and thereby filter for specific modules or pathways of interest.
Metabolite names followed by * indicates a compound that has not been confirmed based on a standard, but Metabolon Inc. is confident in its identity. Metabolite names followed by ** indicates a compound for which a standard is not available, but Metabolon Inc. is reasonably confident in its identity or the information provided.

Lipidomics dataset

The following figures are the results of the WGCNA (Weighted Gene Co-expression Network Analysis) run using lipidomics data from patients hospitalised with influenza or SARS-CoV-2. The lipids are fully quantitative and measured in uM. Missing (not measured) values were imputed using 0.0000001, non-varied lipids were removed and the measurements were log2-transformed to obtain normal distributions with medians of 0.

Modules

Groups of metabolites are found using the WGCNA package (link). First, correlations between all lipids were calculated using the bi-weight mid-correlation measure which is median-based and less sensitive to outliers. Next, the topological overlap of the correlations was found. This measure takes into account both the molecule-to-molecule correlations, but also the two molecules’ shared correlations, which is a way to simulate biological pathway structures. Hierarchical clustering of the topological overlap measures was done to find groups of lipids with similar abundance patterns across the patient cohort. Clustering of the lipidomics data resulted in 12 modules whereof one, the grey module, contains the lipids that did not fit in any of the remaining clusters. All modules are repressented by a colour and can be seen in the dendrogram below where each branch represents a lipids. Keep in mind that this is a 2D schematic of a multidimensional figure and each “hinge” in the tree can be flipped.

Correlations with clinical variables

Spearman’s Rank Correlations were calculated for the weighted average abundances (first principal component) of all lipids in each module association clinical variables. Red indicate a positive correlation and blue a negative one. NB: not all lipids in a cluster necessarily show the same correlation. The asterisks indicate false discovery rate (FDR) adjusted p-values: * = q<0.05, ** = q<0.01, *** = q<0.0001. The clusters are represented by the colour column to the right and the dendrogram indicate distances between the modules.















Super- and sub-pathway content of modules

In the tabs below are figures showing the super- and sub-pathway content of each module created with WGCNA. Hovering over the figure will show which lipids are in each category.
NB: Some modules contain too many lipids to produce a nice figure. The full content of all modules can be found in the table below.

black

purple

greenyellow

magenta

pink

blue

green

yellow

red

brown

turquoise

grey

Table of lipids in modules

All lipids can be found in the table below along with information about the molecule and which module it was placed withing. You can type in search terms and thereby filter for specific modules or classes of interest.