inst/troubleshooting.md

In nathaniel-mahieu/warpgroup: Advanced Alignment, Grouping, and Consensus Peak Bounding of LC/MS Data

Common Questions

• Where is warp.consistency? This parameter is computed when the flag detailed.groupinfo is set to true. Eg. group.warpgroup(detailed.groupinfo=T) It is a per-peak metric found in the peaktable if using XCMS.

• Where is n? n represents the number of originally detected peaks still contributing to the output warpgroup. It can be summarized form the output by counting all the non NA values in xs@peaks[,"n"] per group.

• Memory Issues the algorithm breaks the job up into small parts, the largest part is defined by the group in @groupidx which is most memory intensive. This would be the group which covers the largest RT range. If you filter and remove groups whose peaks min(rtmin) to max(rtmax) span a very large range you will probably fix your memory issues. Maybe plot a histogram and see if you have one super wide group.

• Warpgroup never finishes running. Always test warpgroup on your specific settings and data with a subset to make sure it will complete in a reasonable amount of time. Due to poor scaling it is possible your analysis will never complete. Ex.

xs2 = xs
xs2@groupidx = sample(xs2@groupidx, 10)
start = Sys.time()
group.warpgroup(xs2, ...)
end = Sys.time()
paste("Full dataset will take", length(xs@groupidx)/length(xs2@groupidx), "times as long to process.")
paste("Subset took", end - start)

• XCMS Function xxxxx returns an error Warpgroup returns the output of the algorithm and packs it in an XCMS data structure. Warpgroup does not ensure that everything XCMS expects is true of the data. For example, if no ion intensity was detected for one peakgroup in a sample Warpgroup does not assign a peak m/z or retention time to that sample (there was none). Most known errors are due to NA values in the peak table. Try removing all peaks with NA values and rebuilding the groupvals with something like below.

missing.values = which(is.na(xs@peaks[,c("mz", "mzmin", "mzmax", "rt", "rtmin", "rtmax", "into")]), arr.ind = T)
for (i in seq(nrow(missing.values))) {
  ps = xs@groupidx[[xs@peaks[i,"new.gidx"]]]
  xs@peaks[i, missing.values[i,]] = mean(xs@peaks[ps, missing.values[i,]], na.rm=T)
}

xs = warpgroup::refreshGroups(xs)

• How long will X take The record as far as I know is 216 files over 5 days. This included 567 groups and 2027 scans and a rt.max.drift of 32 scans.

Parallel Computing

Warpgroup supports parallel computation of a job via the foreach backend. Simply mount any foreach compatible backend and warpgroup will automatically use that to distribute the job. doParallel is great for local cores and doRedis is great for distributing the job among multiple machines. For parallel computing something like the following example code should be run before starting group.warpgroup.

doParallel

library(doParallel)
cl = makeCluster(detectCores() - 1)
registerDoParallel(cl)

doRedis

Master node

library(doRedis)
registerDoRedis("worklist.name", "IP.OF.REDIS.SERVER")

Processing node(s)

library(doRedis)
startLocalWorkers(n, "worklist.name", "IP.OF.REDIS.SERVER")

Thank You

Thank you to the following for beta testing and bug report assistance. - Kevin Cho - Robert Sander Jansen

nathaniel-mahieu/warpgroup documentation built on May 23, 2019, 12:19 p.m.