Convert Spectra to array of python Spectrum objects

[jorainer]

I assume the Spectrum class is defined in matchms? is this a general object also used in other packages? @hechth can you maybe provide more info here?

Also, is there a documentation on the additional attributes and the expected data types for this type of spectrum?

[michaelwitting]

The documentation of the Spectrum class can be found here:
https://matchms.readthedocs.io/en/latest/api/matchms.Spectrum.html

[hechth]

The additional attributes are just stored in a dict in the Metadata class - I'm not sure how dicts are managed in reticulate, bu the individual attributes can be accessed.

Maybe we can schedule an initial meeting to discuss the aims of this project? I think that might be more efficient :)

[michaelwitting]

Python dict equals to an R named list

[hechth]

@jorainer The Spectrum object is likely used in Spec2Vec, MS2DeepScore and also in most of our python based tools - the actual peaks are stored in a separate object.

So you want something that takes a Spectra R object and gives you a list of Spectrum python objects?

[jorainer]

Exactly @hechth ! Ideally with an option to set all, or only a reduced set of additional spectra variables we have in Spectra.

And then also the way back would be cool (list of Spectrum -> Spectra).

[michaelwitting]

Super simple solution, without carrying any metadata for the moment:

# function to convert an R Spectra to a Python Spectrum
rspect_to_pyspec <- function(x, reference = import("matchms")) {
  
  reference$Spectrum(mz = np_array(x$mz[[1]]),
                      intensities = np_array(x$intensity[[1]]))
  
}

# convert spectra to python spectrum
spectrum1 <- spectrum2 <- spectrapply(sps, rspect_to_pyspec) %>% unname()

Does any function within matchms, Spec2Vec, MS2DeepScore etc, use any other information, e.g. precursor m/z etc? I guess this has to go to the Metadata, right?

[hechth]

I don't think Spec2Vec and MS2DeepScore do, the MetadataMatch class does

[jorainer]

So, maybe a good solution would be to have a converter function that has a parameter to allow defining which (if any) spectra variables should be converted to Python. I would assume converting just the peaks matrix would be faster than considering also metadata, and if matchms does not use metadata anyway we should also allow to avoid this additional conversion (for performance reasons).

[michaelwitting]

I have in my branch already some functions. I can modify them to have a selection of metadata to be carried along.

[hechth]

Matchms does use metadata - it can be used for filters - since metadata is just holding a dict, any kind of metadata can be stored.