I ran PCA on just the batch QC samples. I also plotted the intensity of several, randomly selected markers across all my QC samples. I've attached a few of the plots.
There does appear to be a pretty consistent increasing trend over the batches (with the exception of batch 20180728). This seems suspicious. I'd expect the intensities to decrease or increase over time depending on when I cleaned the sample cone.
The trend seen in the individual markers does seems consistent with the trend seen in PC1, though.
Also, I'm wondering what the lower intensity points are. Perhaps they were generated during the peak filling step. I'm not sure. If that's the case, maybe my bw=1 in PeakDensityParam is--as you suggested--too low?
These are not known compounds as you suggested. They they are just randomly selected markers. My reasoning was that all markers are contributing to the PCA. However, it does seem wise to quantify a few peaks using QuanLynx and see if the trend matches with what I'm seeing in the markers generated by XCMS.
Btw, here's the code I used to generate the plots of individual markers:
library(dplyr) # for sorting the dataframe
dfr <- arrange(dfr, batch) # sort the dataframe by batch
m <- dim(dfr)[2] # number of columns in the dataframe
rand <- function() {round(runif(1, 16, m), 0)} # function to generate a random column number for indexing; random integer between 16 (i.e. the first column of features) and the last column
plot.it <- function(){ # create a function to simplify plotting as below
n <- rand()
plot(dfr[,n]
, col=dfr$batch
, main=c("marker #", (n-15), "batch QC samples only")
, pch=20
, ylab="peak area"
, xlab="sample number"
) # end of plot call
legend("topleft"
, levels(dfr$batch)
, col=1:length(levels(dfr$batch))
, cex=0.8
, pch=20
, title="Batch"
)# end of legend call
} # end function definition for "plot.it"
plot.it() # just call this to generate plot of a new random marker
[attachment deleted by admin]