MS2DeepScore

[eeko-kon]

Hi Niek!

Just an enhancement suggestion for your tool: as a user, since I am already using MS2DeepScore, it would be nice to get a table that I could use for networking with the nice annotations and classifications instead of having to re-run MS2DeepScore. :) Just a future suggestion!

[justinjjvanderhooft]

Thanks for your nice suggestion Efi!

[niekdejonge]

Hi Efi,

Great suggestion, I agree!

Something that is very easy to add is the ms2deepscore between your query spectrum and the predicted analogues (or actually already option I believe). However, this is not useful for networking.

Instead you would need the ms2deepscores between your query spectra.
This requires a bit more redesign of the code, but is certainly doable.

Benefit:

• The model is already downloaded
• Code for running MS2Deepscore is already there
• The embeddings for the query spectra are already calculated (longest step of MS2Deepscore)

Downside:

• Not really the main purpose for MS2Query, we have to watch out for too much code complexity.

It is certainly possible to implement in MS2Query, but would need some rewriting, since we currently do not store the query ms2deepscore embeddings.
If implemented, it would be great to also add a tutorial for molecular networking with MS2Query.

Notes to myself/design proposal:
Store ms2deepscore query embeddings in MS2Library. (in _get_alL_ms2ds_scores).
Add a method to ms2library which runs MS2Deepscore between query spectra and returns a csv file with a matrix with the scores. (check required input GNPS). Maybe use cutoff >0.8 already?
How to note which query spectrum is which? Use the Feature Id from feature based molecular networking? This can than be selected from both the analogue search results and MS2Deepscore matrix to generate a mol network with annotations. Again format in the same way of GNPS so it can be used directly.
Do we want to create one mol network over multiple files? Or maybe just use the MZMine FBMN output. So that would be one mgf file with features and a feature file which stores which files contain these features. Over multiple files is also possible, but this could be a bit more complex to implement and will be hard to denote which spectrum comes from which file?

[eeko-kon]

I totally understand the downside- this could possibly be a different tool/version/repository.

I would say the main point would be to create an output that could be used in Cytoscape. This could easily be a table with the relationships between the features, as you mentioned, through feature_IDs that are unique or, as FBMN does, generate a new ID that can still be linked back to the MGF file somehow (but that is not necessary).

If you need any help with the mol networking tutorial or figuring out the table details I am happy to help!