Tutorial - How to use LINEX

 1. Upload

In order to use LINEX a Lipid data file needs to be uploaded in .csv format (Comma separated values) or .tsv format (Tab separated values).
Additionally, a file containing Sample Groups, Lipid Class Settings or Fatty Acid Settings can be uploaded.
Structure of the files:

The uploaded files should follow the scheme shown in Section 1.1. An example lipidomics dataset can be used for testing and exploring.

1.1. File formats
Lipid data file
Lipid1 Lipid2 ...
Sample1 1.2 2.3 ...
Sample2 3.4 4.5 ...
... ... ... ...
Sample group file
Sample1 name1
Sample2 name2
... ...
Lipid class settings
>Class Connections
PC: PE, DG, ...
PE: PC, DG, ...
Class FA Numbers
DG: 2
TG: 3
Fatty acid settings
Fatty Acids
Classes: SM, Cer, ...
Excluded Reactions
20:3, 20:4
Elongation: C: 2, DB: 0, OH: 0
1.2. Computation Options

The default options can be used as a standard. To optimize runtime make sure only the required checkboxes are selected.

  • Check if the input data is log-transformed. This is important for correct fold-change calculations.
  • Check if the computed fold-changes should be log transformed. It has no effect if the data is already log-transformed (see above)
  • It should only be checked if lipid species names are not yet in the LipidLynxX nomenclature. Be aware that converting names may take up to several minutes. Re-conversion of already converted species increases runtime and might throw errors for molecular species.
  • If you need to convert, please make sure you are following one of the supported notations (see lynx github at 'Supported lipid notation styles')
  • To avoid the conversion process multiple times for the same file you can download the converted data ('LipidLynxX Converted Data as .csv') once the network computation has finished. A list of lipid species with incorrect notations will also be available ('Unconverted Lipid Species') on the download page.

Lipid Resolution

  • Setting the highest resolution to consider, this means all lipids with a lower resolution are considered as they are. If you set the highest resolution to 'Molecular Species', for example, sum species will be kept as such, but sn-specific annotations will only be considered as non sn-specific. Please only change this parameter if all lipids are sum species or some are confident sn-specific identifications.
    Ordered in decreasing resolution: Sn-specific > Molecular Species > Sum species.
  • Check if Correlations/Correlation Changes should be computed. Missing values just skipped.
    See node colouring options (Section 3) for a short explanation for how correlation changes are computed and what they mean.
    Unsignificant correlations (FDR > .05) are automatically set to 0.
  • Check if Partial-Correlations/-Correlation Changes should be computed. Only possible when there are no missing values.
    Partial correlations only consider 'direct' correlation, i.e. without the effect of all other lipids, in contrast to 'regular' correlations which also show indirect relations.
    See node colouring options (Section 3) for a short explanation for how partial correlation changes are computed and what they mean.
    Unsignificant partial correlations (FDR > .05) are automatically set to 0 if p-values (from Fisher's z-transform) could be computed. If the sample to feature ratio is too small p-values cannot be computed and you will be notified by a warning. In this case all partial correlation values are displayed as they are.
  • Check if (log) Fold Changes should be computed. Fold changes are computed as groupA/groupB or groupA - groupB respectively.
    Positive values indicate higher values in the group named at first position!.

Reference Group

  • If you uploaded Sample groups and no Reference Group is chosen all pairwise combinations between sample groups will be calculated.
    If there are Sample groups uploaded and a Reference Group is given fold-changes, p-values and (partial) correlation changes are computed against the Reference Group.
    If did not upload Sample groups this will have no effect.
    Please ensure correct spelling (heading and trailing whitespaces are ignored).
  • Check if p-values (binary statistical tests) should be computed.
    All p-values are automatically FDR correct using the Benjamini-Hochberg procedure.
    If the data is approximately normally distributed t-test should be used to get more powerful results. Otherwise we recommend using rank-based tests (Wilcoxon rank-sum and Mann-Whitney U rank test). In the case of paired data, we recommend using Wilcoxon's signed rank test.
  • Check if the displayed network should be directed. This has no effect on statistical measures, but will lead to reaction and enzyme annotation being assigned to a defined reaction direction.
    The directed graph option will consume a considerably higher amount of memory and cpu!
1.3. Upload & compute network

Check if everything is set correctly and press the button. The computation will start.
The uploaded data can only be used once. If you want to change a Computation option or open a new tab make sure the input data is still selected otherwise it has to be re-uploaded.
Deleting your browser cookies while using LINEX will also lead to a loss of results, because your data is matched by an anonymous session id.
NOTE: if you upload a new dataset all prior results will be lost!

how it looks like during the network computation
When the computation is done a 'View results' button will appear taking you to the analysis page.

 2. Analysis

how it looks like during the network computation

2.1. Network Visualisation Parameters
  • Node Colours
  • Edge Colours
  • Node size
  • Comparison: defines which comparison (e.g. groupA vs. groupB) will be used for FDR-values, fold-changes and (partial) correlation changes.
  • Group: defines which group (e.g. groupA) will be used to display (partial) correlation values.
  • Find Lipid Species
    • Find Lipid species by its name
  • Find by Substring
    • Find lipid species by substrings. Supports Regular Expression (see examples below).
    • PC: will find all lipid species with "PC" in their name (e.g. LPC, PC, PCO etc.)
    • ^PC will find all lipid species whose name is starting with "PC"
    • ^PC|^PE will find all lipid species whose name is starting with "PC" or "PE"
    • ^PC.*16:0 will find all lipid species whose name is starting with "PC" and contains a 16:0
  • Shown reaction types: Select, if all reactions, only lipid class reactions or only fatty acid reactions should be shown.
  • Enable physics: If enabled, the network layout is computed dynamically. This means, that moving one lipid with the cursor will move the whole netowrk dynamically.
2.2. Legend
  • Node colours (only)
    • Lipid Classes
      • Lipid class colours are pre-defined (see . For a description on how the colour map was generated refer to the LINEX publication.
      • If you add custom lipid classes not covered by our colour scheme, it will default to a dark grey.
      • It is currently not possible to choose specific lipid class colour maps
  • Edge colours
    • Lipid Reactions: kind of reaction an edge is representing
      • Chain length: change of the number of C atoms
      • Desaturation: change of the number of double bonds
      • FA addition: connection of two lipids with the same head group but different numbers of fatty acids
      • Head Group Modification: head group related reaction between two lipid species with the same fatty acids/sum composition
      • Attention: when customising fatty acid reactions in which more than one FA property is changed at ones colour annotation will not be accurate anymore
    • Correlations: correlation between the connected lipids
      Shown correlations are statistically significant
    • Correlation Changes: discrete categories of changes in correlation between two sample groups.
      'significant' and 'unsignificant' refer to statistical significance
      • negative to positive: significant correlation in both groups, <0 in groupA and >0 groupB
      • positive to negative: significant correlation in both groups, >0 in groupA and <0 groupB
      • significant to unsignificant: significant correlation in groupA, unsignificant in groupB
      • unchanged significant: significant in both groups, either both >0 or both <0
      • unsignificant: unsignificant correlation, i.e. uncorrelated, in both groups
      • unsignificant to significant: unsignificant in groupA, signicant in groupB
    • Partial Correlations: partial correlation between the connected lipids
    • Partial Correlation Changes: discrete groups of changes analogous to 'Correlation Changes'
  • Node size or colour
    • -log10(FDR)
    • Fold Changes
      • Fold changes are computed a groupA/groupB (or groupA - groupB if log-transformed/log-ratios)
      • If the selected comparison is groupA_groupB positive (log) fold-changes indicate higher values in groupA!
    • Desaturation
    • Degree (number of edges per node)
    • C Index
    • Chain Length
    • Betweenness Centrality
      • The Betweenness Centrality quantifies the importance of a node. It indicates how many shortest paths between any couple of nodes in the graphs pass through a node.
    • Closeness Centrality
      • The Closeness Centrality indicates the accessibility of a node. It is a measure of the average shortest distance from a node to all other nodes.
    • DB Index
2.3. Additional Features

By clicking on nodes or edges in the colour legend all unselected edge/node types will turn grey in the network. Multiple selection by holding ctrl while selecting is possible. To restore all colours simply click on a non-populated spot inside the colour legend area.
The size legend will be scaled according to the network zoom (i.e. re-scaled every time the network is zoomed in or out). For better readibility it is also possible to zoom the size legend individually.
Zooming via mouse scroll and moving nodes with the cursor is supported for both the network and legend elements. Additional information can be viewed by hovering over edges and nodes. The current network view can also be downloaded as a pdf file.

 3. Download

Choose the desired files to download
how it looks like during the network computation

3.1 Network related files

The network can be downloaded as .html file or as .graphml.
The standalone html file can be opened in a browser (locally) and resembles the analysis page with all interactive functions.
Graphml is a general graph storage format. It can be used to import the network into graph tools like Cytoscape in order to generate publication ready plots.
By downloading colour and size legend data the same tools can also be used to create the corresponding legends in a suitable format.
To show the lipid class connections used in your session you can download the corresponding graph as a png image or as a graphml file.

3.2 LipidLynxX related files

LipidLynxX Converted Data along with a list of incorrectly annotated species names can only be downloaded if the conversion options was checked on the upload page.

3.3 Statistical Metrics

All computed statistics (as defined on the upload page) can be downloaded as csv files. If sample groups were given one file for each group (or comparison) will be provided.
For (partial) correlations the full correlation matrix is returned, thus also including values for lipid pairs with no (direct) connection in the network.

 4. Delete Data

Data is saved on the server for 100 minutes after upload and will automatically be deleted afterwards, but it can be deleted manually here as well.