In MetClassNet/MetClassNetR: Create and analize networks built from metabolomics Data

knitr::opts_chunk$set(echo = TRUE)

library(MetClassNetR)
library(MsCoreUtils)
library(tidyverse)
library(dplyr)
library(QFeatures)

Buildig a multi-layer network for C. elegans

This vignette demonstrates on example data from C. elegans the use of MetClassNetR to generate a multilayer network.

Stepwise it is shown how three experimental networks are build from experimental MS data, which is given in maf format. In a next step the experimental networks are connected to the Genome-Scale Metabolic Network (GSMN) of C. elegans, published in WormJam, using the tool Metabolomics2Network (https://forgemia.inra.fr/metexplore/metabolomics2network). Metabolomics2Network is a Python package which aims to map the assigned ChEBI IDs from the experimental feature list to the ChEBI IDs of the GSMN metabolites based on connections within the ChEBI ontology (https://doi.org/10.1093/nar/gkv1031). If an exact match is found, the reported distance between feature and metabolites is zero. It will be be larger and lower than zero, depending if the metabolite is mapped to a more general metabolite (parent class) or to a more specific metabolite (child), respectively.

Please note that Python is required as well as the following Python packages: - Metabolomics2Network - libchebipy - urllib3 - Sphinx - sphinxcontrib-napoleon

Load input data

The first step in this workflow is to load the input data, using the loadInputData function.

The output of the loadInputData function is a named list containing all the
data (i.e., peakList, spectra, transformations, and gsmn).

# load("../VariablesFromVignette.RData")
path <- "extdata/MTBLS1586/"
exp <- "MTBLS1586_LC-MS_positive_reverse-phase_metabolite_profiling"

inputData <- 
  loadInputData(
    peakListF = 
      system.file(
        paste0(path, "m_", exp, "_v2_maf.tsv"), 
        package = "MetClassNetR"
        ), 
    intCol = 23, 
    transF = 
      system.file(
        paste0(path, "transformations_MTBLS1586.csv"), 
        package = "MetClassNetR"
        ),
    spectraF = 
      system.file(
        paste0(path, "ms2_", exp, ".mgf"), 
        package = "MetClassNetR"
        ), 
    gsmnF = 
      system.file(
        paste0(path, "WormJam-GEM-20190101_L3_no-side_no-comp.gml"), 
        package = "MetClassNetR"
        ),  
    resPath = "~/MetClassNetR_MultiLayerNetwork/", 
    met2NetDir = 
      paste0(
        find.package("MetClassNetR"), 
        "/Python/metabolomics2network-master/"
        ),
    configF =  
      system.file(
        paste0(path, "Metabolomics2NetworksData/JsonConf.txt"), 
        package = "MetClassNetR"
        ),
    idenMetF =  
      system.file(
        paste0(
          path, 
          "Metabolomics2NetworksData/IdentifiedMet_", 
          exp, 
          "_v2_maf.tsv"
          ), 
        package = "MetClassNetR"
        ),
    metF = 
      system.file(
        paste0(path, "Metabolomics2NetworksData/WormJamMetWithMasses.tsv"), 
        package = "MetClassNetR"
        ), 
    spectraSS = 50,
    cleanMetF = FALSE
    ) 

Build experimental networks

The second step is to build up to three experimental networks, based on spectral similarity, correlation, and mass difference. We build the experimental networks using the buildExpNet function. This will take some minutes depending on your system.

The output of the buildExpNet function is a list of experimental networks as igraph objects.

expNetworks <-
  buildExpNet(
    inputData, 
    net2Build = "all", 
    directed = FALSE, 
    ppmMass = 10, 
    ppmSpec = 0,
    tol = 0.005, 
#    corrModel = "pearson_partial",
    corrModel = "pearson",
    corrThresh = 0.25
    )

Map identified metabolites to the GSMN

The third step is to map the manually identified features to the corresponding metabolites in the GSMN, using the ChEBI id of such features. Notably, although it is possible to do exact matching and potentially have a low covering (i.e., to have very few matches since not all the metabolites participate in known chemical reactions of a GSMN), we could also use ontologies (such as ChEBI) and maximize the covering.

## Metabolomics2Network
resFile <- "Res_Met2Net_MappedMet.txt"

mapMetToGSMN(inputData, resFile, method="metabolomics2network")

Finally, we are ready to finish building the multi-layer network, joining the experimental networks with the GSMN, thanks to the mappings we just obtained. To this end, we call the makeMultiLayer function with the inputData object, the experimental networks (stored in expNetworks), and the path to the folder that contains the mappings, which is called GMSNMappings, and it is located inside inpuData$resPath.

In order to visualize and/or save the multi-layer network as two files: one containing the list of edges and another one with the list of nodes, we call the writeMultiLayer function with the inputData and multiLayer objects, and a boolean value for the parameter visualize that should be TRUE if the multi-layer network is to be visualized in Cytoscape or FALSE otherwise. Please be aware that Cytoscape should be open if visualize == TRUE.

pathToMappings <- paste0(inputData$resPath, "GSMNMappings/")

multiLayer <- makeMultiLayer(inputData, expNetworks, pathToMappings)

writeMultiLayer(inputData, multiLayer, visualize = FALSE)

Analyzing the multi-layer network

As a final step, we will analyze the interconnection level between the experimental layers and the GSMN of the multi-layer network. For that, we will calculate some basic stats, using the calculateMultiLayerStats function, which generates some plots and stores them in the resPath directory (defined in inputData.

calculateMultiLayerStats(multiLayer, inputData)

sessionInfo()

MetClassNet/MetClassNetR documentation built on June 30, 2023, 2:12 p.m.