Input summary
Missingness by group
Before filtering, with technical samples still included. A group that is much better detected than the rest — pooled QCs usually are — will pull the detection probability curve if it is left in the model.
Reading this table. N is samples in that group. Missing % is over every cell of the group's block of the matrix; Median per-sample missing % is the typical single sample, so a big gap between the two means a few very poor samples are carrying it. Proteins detected counts rows seen at least once anywhere in the group — near the total means nothing is uniquely absent there. Samples kept is how many survive the technical-sample exclusion in section 3. A count short of N means some QC or blank injections were labelled with that group's value and were dropped — not that the group itself is excluded. The All row is every sample pooled.
Sample matching
How each DIA-NN run column was matched to a row of the phenotype file.
Missingness after filtering
Intensity distribution by sample
Post-normalisation log2 intensities; look for samples sitting off the common centre.
Per-sample detection
Detection probability curve
The DPC gives the probability that a precursor is detected as a function of its underlying log2 intensity. A steeper slope means more information can be recovered from a missing value.
Average intensity vs missing values
Mean-variance trend (vooma)
Protein-level standard errors from dpcQuant() become precision weights for the linear model.
Sample quality weights
Principal component analysis
Scree plot
limpa MDS (uses quantification standard errors)
plotMDSUsingSEs() downweights proteins that were quantified imprecisely.
Sample-to-sample correlation
Volcano plot
P-value distribution
A flat histogram with a peak near zero is what you want. A peak near one suggests the model is over-conservative — try turning sample weights off.
MA plot
Top proteins
Rows are z-scored; samples are ordered by group.
Gene set enrichment (fgsea)
Proteins are ranked by -log10(p) x sign(logFC) on the raw p-value, then tested against MSigDB collections. This runs on the current contrast.
Top pathways by normalised enrichment score
Enrichment curve
Newest messages last.