That's an interesting statement when it was recently shown in multiple publications (e.g. www.nature.com/articles/s41...) that DIA-NN v1.8.1 (superseded by v1.9 shortly after ASMS 2024) does not consistently control FDR. Great to see that more recent versions seem to have improved here.

Tobias finds that DIA-NN 1.9.2 has FDR better under control than 1.8.1 for his dataset, but Spectronaut 19 seems to still be off.

Glad to see this published. I hope its message reaches both the communities of practitioners that want to simply use and be able to trust software for the analysis of DIA data, as well as the developers of said software. This is important for the field!

Please see my replies in the two posts linked below for clarification.

bsky.app/profile/mart...

and

bsky.app/profile/mart...

We believe the curves look like this for Spectronaut and DIA-NN because the limited search space of the 2nd pass search obviates proper run-specific FDR control.

On platform.msaid.io, we already support semi-tryptic searches. Not quite HLA (yet), but I'd suggest to stay tuned!

Please see my replies in the two posts linked below for clarification.

bsky.app/profile/mart...

and

bsky.app/profile/mart...

We believe the curves look like this for Spectronaut and DIA-NN because the limited search space of the 2nd pass search obviates proper run-specific FDR control.

We believe the ID increase is a function of the sensitivity of the instrument. But ultimately, comparisons like this one are always limited by the fact that one cannot really control for all parameters.

Exactly, because identifications with an eFDR > 1% cannot be called identified at an FDR <= 1%.

You are missing the point. This investigation focuses on the correlation of run-specific precursor-level q-values and run-specific precursor-level entrapment q-values. When reporting a list of IDs, this list has to be controlled at the corresponding FDR, which is what we plotted.

We included the time to generate a spectral library, because spectral libraries are often project-specific and in our experience are rarely generate only once in practice.

We did not change the plots, because Q.Value is the run-specific precursor q-value according to github.com/vdemichev/Di.... This x-axis is the same for all softwares we compared. You can find your suggested version of Extended Data Fig. 6B below. How do you interpret it?

That post comes from a time when the Tribrids and the Hybrids were not comparable in terms of NCE.

Chimeric DDA spectra are only 'junk' if you can't deconvolute them properly in a spectrum-centric fashion. If only there was a software capable of doing that... 😅 Check out our preprint on solving this problem: doi.org/10.1101/2024.... DDA, DIA and PRM data analysis with the same algorithm!

We are of course using INFERYS, but yeah. 😅 Our verification is the increase in spectral angle.

Given the data we see, I believe Thermo also spent quite some time on making sure that Astral spectra are similar to the spectra from Hybrids and Tribrids. Maybe @hamishs.bsky.social wants to chip in here.

It might also be related to the actual function that maps CE to NCE. In the past, spectra of the same peptide acquired at NCE 25 on the Tribrids and Hybrids from Thermo looked different. I hear that has since been addressed in an effort to make spectra comparable.