Too low number of IDs with NNs: reverting to linear classifer

[joycelyny]

We have a dataset with two groups that have empty results files when opening the report.tsv, and we think it may be because the linear classifer was used. Is there a way to override this manually?

[vdemichev]

Hi,

How does the log look like?

Best,
Vadim

[joycelyny]

[vdemichev]

Likely means nothing is detectable in that run.
What can increase ID numbers is including common contaminants, but it will not really help much in identifying the proteins of interest in that run.

[joycelyny]

I see! Are there suggested modifications to make within the DIA-NN workflow to prevent this? I have multiple experiments that seem to have the same problem.

[vdemichev]

That's most likely the problem with runs, not with DIA-NN. But can try with --mass-acc-cal 100 (and a DIA-based empirical spectral library), this will solve the problem if calibration is the issue.
XICs look OK, everything else seems OK?