Integrate functionality from other Python libraries for MS data analysis

[jorainer]

This is related to issue #18 but focused on functionality from other Python libraries.

Discussion: which libraries could be added? Is there need for integration of these into R-based workflows? What would be the use-cases?

[mdgrv]

What are the most important MS-related Python libraries used nowadays? Same question goes for the most important MS-related R packages? E.g., packages/algorithms tackling pre-processing, annotation, ... of MS. We can focus on which have a broad user-base and make those easily interchangeable between R and py.

[bittremieux]

Relevant Python packages:

• matchms
• spectrum_utils
• Pyteomics
• pyOpenMS
• pyOpenMS-viz

[jorainer]

Thanks! but I assume they all use different classes to contain the MS data? Would be cool if SpectriPy could translate Spectra objects into any of these - or at least let's start with spectrum_utils.

[bittremieux]

Yes indeed. Interoperability between Python is unfortunately a separate problem. We can indeed start with matchms and/or spectrum_utils.

[jorainer]

@bittremieux , I just added a simple example how the full m/z and intensity values of a whole Spectra object could be extracted - and then accessed in Python (since it's just a list of numerical two-column arrays). See https://github.com/rformassspectrometry/SpectriPy/blob/main/inst/doc/cross-language-ms-analysis.qmd#L135-L155 - if you could provide some Python code to (efficiently) get that into a spectrum_utils.MsmsSpectrum (assume a list of these objects?) I could write the related R code/function in SpectriPy - and we would have already a first support for spectrum_utils :)

[bittremieux]

Question is also which other spectra variables (metadata) is required or supported by spectrum_utils.<

It would look something like this:

## su <- spectrum_utils import

pest_ms2 <- filterMsLevel(pest_dda, 2L)
pd <- as.list(peaksData(pest_ms2))
precursor_mz <- precursorMz(pest_ms2)
precursor_charge <- precursorCharge(pest_ms2)

## Some R magic to iterate over each spectrum needed here.
spec <- su$MsmsSpectrum(
    "spec",             # random spectrum id, or the actual id if the Spectra have them.
    precursor_mz,
    precursor_charge,
    np_array(pd[, 1L]),
    np_array(pd[, 2L])
)

But Thomas might have gotten pretty far with this as well.

[jorainer]

Ah, that's great Wout! Thanks. no, indeed we were not sure how metadata could be passed to MsmsSpectrum (#47 )

[tnaake]

For clarification: this is how the MsmsSpectrum is constructed (link to code)

res <- reference$spectrum$MsmsSpectrum(mz = np_array(pks[, 1L]),
            intensity = np_array(pks[, 2L]), 
            identifier = "Spectra_to_MsmsSpectrum", 
            precursor_mz = precursorMz(x), 
            precursor_charge = precursorCharge(x))

Indeed, we might loose some information stored in Spectraobjects, but we can think of adding them after the conversion, i.e. once we convert back from MsmsSpectrum -> Spectra, to the Spectra object.

[bittremieux]

The benefit of Python is that any additional attributes can be assigned without needing to be defined in advance. I often add custom metadata to spectra. I'm not sure whether reticulate also supports it, but I'd assume it should. In which case you would do something like this:

res <- reference$MsmsSpectrum(...)
res$ms_level <- msLevel(x)

[tnaake]

Thanks for the clarification, @bittremieux, I will look into this :)