Thank you for your comments. Please note that I did not apply any filtering to the peptide intensity. The experimental protocol includes two groups (WT vs. KO), with four replicates per group. All samples were acquired on the same day using an Astral spectrometer. I did not perform the DIA-NN myself and have no experience in adjusting DIA-NN parameters.

As you reported outliers, I performed a limma::plotMDS. The first figure shows the top 5,000 peptides (out of 95,000) and the second shows the 5,000 peptides with the highest mean. Both figures highlight that the KO group is not homogeneous, unlike the WT group. I am currently investigating this point.

I studied the distribution of the mean intensity relative to the number of non-missing values as it was not clear to me on the plotDPC. My code is at the end in order to check I did a msitake. Each distribution was normalized to its highest value (i.e. by line on the heat map). The complete dataset and the dataset divided by group are shown below. There is a slight shift between the KO and WT groups, but I am puzzled by the clear shift of 1 to 2 log2(intensity) between no missing values and one or more missing values. Is this usual? The value missingness seems to be quite independant of the missing value count.

Regarding the slope of the DPC curve, I wonder if it could depend on the characteristics of the spectrometer. Have you worked on an Astral dataset?

# heatmap of mean value distribution as a function of non-missing value count
# ALL samples
res <- list(
  as.data.frame( t( apply( y.peptide, 1, function(x) {
    c( detected = sum(!is.na(x)), mu_obs = mean( x, na.rm = TRUE ) )
  })))
)
names(res) <- "all"
# PER GROUP
res <- lapply( group_levels , function( glv ) {
  as.data.frame( t( apply( y.peptide[ , y.peptide$targets$group == glv ], 1, function(x) {
    c( detected = sum(!is.na(x)), mu_obs = mean( x, na.rm = TRUE ) )
  })))
})
names(res) <- group_levels
# heatmap plot (same for ALL and PER GROUP)
for ( glv in names(res) ) {
  dat <- res[[glv]]
  tbl <- prop.table(table( 
    dat$detected, cut( dat$mu_obs, breaks = seq( 8, 38 ) ) ), margin = 1)
  tbl <- tbl[ setdiff(rownames(tbl), "0"),]  # remove 0 count
  tbl <- tbl[ nrow(tbl):1, ]  # reverse order for display
  pheatmap::pheatmap(
    as.matrix( tbl ), cluster_rows = FALSE, cluster_cols = FALSE,
    main = glv)
}