Documentation on spectrum / RT / IM prediction in library free mode

[IwanParf]

Hello Vadim, you have built a great piece of software, which is of tremendous help, thank you and congrats to that! I was just wondering about how DIA-NN is predicting spectra for the library-free mode. I could not find a documentation on that topic in the 2020 Nature Methods paper, although in the supplements it is mentioned and refers to the methods section (Chapter 8, library-free processing: "proteins are in silico digested and prediction of the fragmentation spectra of the resulting peptides as well as the respective retention times is performed (Methods)"), where, again, I could not find the info.

Best regards,

Iwan

[vdemichev]

Hi Iwan,

The exact algorithm is not published, but this is similar architecture based on recurrent neural networks as in other works like Prosit.

Best, Vadim

[IwanParf]

Hey, thank you for the response! Is it planned to publish that work? I frequently work with data coming from DIA-NN and when I explain the analysis steps done on stakeholder data, it would be helpful to somewhat being able to explain what DIA-NN does (excluding the black box problem in ML, of course). But thanks anyway!

Iwan

[lazear]

Hi Vadim,

I would like to second Iwan's request for more information here. It would be great to know how the predictor works! Unfortunately, it appears that even going back to when DiaNN was open source, the code for the Predictor class/project was never checked into the repository 😢

Best, Mike