search output library

[JM-Bader]

One can create a new spectral library as part of a DIANN search.

Q1: What is the envisioned application case of that?
Q2: If the search was started with an input library and produces an output library, how do they differ? Does the new library only contain the precursors identified in the search?
Q3: Sometimes directDIA (ie. library-free DIA) can be slow. Is it legitimate to run such a library-free search on a subset of samples to generate a library and use this library in a second search on the full set of samples ?
Q4: DIANN has the Speed and RAM usage options which are kind of an alternative to the suggested approach in Q3. Q3 should be better than compromising via these settings, isn't it?

[vdemichev]

Q1: (i) creation of libraries from DIA data, (ii) this is done automatically during MBR.
Q2: yes, only precursors at 1% global FDR, or whatever FDR filter is set; what info is stored for these precursors is determined by the 'Library generation' setting.
Q3: Yes, just make sure the subset of samples used is representative of the whole dataset.
Q4: Running on a subset should have no performance penalty if you use it on hundreds of runs. Amoung speed options only ultra-fast is a game changer, but it significantly reduces IDs and inflates FDR a bit, so not recommended.

[JM-Bader]

I meant something might be wrong with how the script used to visualise ID numbers filters the final report.

Should the final (main) report be filtered, when I want protein-level information (basically pg.matrix information) ?

[JM-Bader]

Can you please repeat a comparison using the main report? Based on what you showed previously, the nubmers you plotted are fully expected.

Yes, will do. I understand that this is the proposed solution. I just dont fully understand the context of it yet.

[vdemichev]

When searching with a DIA-based lib, that's what you do during research, regardless MBR or not, the numbers in pg_matrix are SEVERELY (in case of your runs) reduced because of too strinct filtering.

[vdemichev]

This is by design

[JM-Bader]

it works: there is no depth loss when using the first pass results from the main .tsv of the Re-Search (compared to the second pass results of the initial search):