Confusion Regarding Order of Conditions When Using mummichog

[nikkrae]

Dear all,

I am using the MetaboAnalystR and want to compare two conditions using a table of peak areas. However, independent of the order of the columns that contain the values for each sample, the results are the same. Shouldn’t the order of the samples affect the pre-fix of the change/enrichment, e.g., condition A vs condition B is positively enriched in mechanism x, while condition B vs condition A is negatively enriched in mechanism x. Is there any argument that reorders the columns prior to the analysis, independent of the order beforehand? I am looking forward to hearing your feedback.

Thanks!

##Mummichog analysis on MS1 peak table

# Create objects for storing processed data
mSet <- InitDataObjects("mass_table", "mummichog", FALSE)

# Set parameters, ppm is 4.3 here
# Only positive mode (ESI+) included
mSet <- SetPeakFormat(mSet, "mpt")
mSet <- UpdateInstrumentParameters(mSet, 15, ion, "yes", 0.05);
mSet <- SetRTincluded(mSet, "minutes")

# Read Peak table
mSet <- Read.TextData(mSet, paste0("analyst_input_", name, ".csv"), "mpt", "disc");
mSet <- SanityCheckMummichogData(mSet)

# Replace minimum value and data filtration
mSet <- ReplaceMin(mSet)
mSet <- SanityCheckMummichogData(mSet)

mSet <- FilterVariable(mSet, "none", -1, "F", 25, F)

# Perform data normalization
# The normailization distribution will be displaced in figure "norm_0_dpi72.png" and "snorm_0_dpi72.png"
# in your working directory
mSet <- PreparePrenormData(mSet)
mSet <- Normalization(mSet, "NULL", "LogNorm", "NULL", ratio=FALSE, ratioNum=20)

mSet <- PlotNormSummary(mSet, paste0("norm_", name, "_"), "pdf", 72, width=NA)
mSet <- PlotSampleNormSummary(mSet, paste0("snorm_", name, "_"), "pdf", 72, width=NA)

# Perform functional analysis with mummichog algorithm
mSet <- SetPeakEnrichMethod(mSet, "integ", "v2") #Alternatively, set method to "GSEA" or "mum"
mSet <- PreparePeakTable4PSEA(mSet)

mSet <- SetMummichogPval(mSet, 0.05)
mSet <- PerformPSEA(mSet, "hsa_mfn", "current", 3 , 100)

mSet <- PlotPeaks2Paths(mSet, paste0("peaks_to_paths_", name, "_"), "pdf", 72, width=NA)

[jeff.xia]

It seems OK to me. You are selecting signficant peaks based on p values (0.05) for enrichment analysis using mummichog - which tests the pathway hits from using the signifcant peaks (say 300 peaks) vs. the pathway hits from 300 peaks randomly selected from the complete peaks (say 3000 peaks). No directions (A/B and B/A) are considered in the computation.

Overall, directions in metabolic pathways are not very widely used as compounds can be substrate or product in a reaction. However, you can always visualize the patterns of pathway abundance profiles across the conditions. See Fig 5 in our Nature Protocol

[nikkrae]

Thank you for your reply, I appreciate it. You are of course correct, the enrichment analysis would not account for a direction of change. However, I have learned that MetaboAnalyst also provides the enrichment methods “GSEA” and “integ” as arguments to the SetPeakEnrichMethod function. Both methods return a Normalised Enrichment Score that shows both negative and positive values. Coming from RNA-Seq, I thought that in this case the order of the conditions makes a difference. However, I am still trying to find the data frame with the results that shows the NES for each pathway identified, as the function only labels the top few pathways. Do you have any insight into this?

[jeff.xia]

Not sure this is possible in the integrative mode. However, you should be able. to do this using GSEA alone - i.e., set the algorithm to be “GSEA”, instead of “integ”