Before I further finalize the table and I add the tests, I have some questions @CunliangGeng. You can already see the results table by running the app from this branch.

• What is the cutoff value we want to have by default for METCALF? Now it's 1. [SOLVED]
• Do we want to display items with a cutoff value higher, lower, or equal to the value set in the scoring filter? Now it's >=. [SOLVED]
• Which of the originally designed columns do we want to include (see the image at the top of this issue)? If they're are not a lot, I could avoid implementing the "Select column" button for now.
• Which of the original hyperlinks in the table do we want to have now (see the image at the top of this issue)? For example, for the "Score" column, it doesn't make sense now to have a hyperlink since we can have only one scoring method.

Let me give my 5 cents here:

• the default cut-off could be 0.05 - that would leave out the most obvious mismatches (and users can always put it on 0.00 if they think that helps their analysis)
• I would go for items equal or higher than the specified value (as you have it now I think)
• these 5 columns (still) look useful to me
• true, though don't we have the regular and normalized Metcalfe score? Anyways, if we have one value, no hyperlink is needed. Let's see if we can add the normalized correlation score asap, maybe as a part of the workshop we could make a start during the parallel sessions.

Thanks for the great work @gcroci2 :-)

• the default cut-off could be 0.05 - that would leave out the most obvious mismatches (and users can always put it on 0.00 if they think that helps their analysis)

Updated to 0.05

• I would go for items equal or higher than the specified value (as you have it now I think)

Indeed :)

• these 5 columns (still) look useful to me

About the possible columns, what are "Product", "Predicted BGC Class", and "Taxonometry"? Not in general, but I mean within NPLinker code-base. So far I have inserted in the table these columns: GCF ID, # Links, Top 1 Spectrum ID, Average Score. @CunliangGeng

• true, though don't we have the regular and normalized Metcalfe score? Anyways, if we have one value, no hyperlink is needed. Let's see if we can add the normalized correlation score asap, maybe as a part of the workshop we could make a start during the parallel sessions.

Currently, for each GCF ID with n links, I calculate the average of the Metcalf scores across all n links.
What is the difference between the regular and normalized Metcalf scores? @justinjjvanderhooft

I've noticed that LinkGraph objects (from npl.get_links(npl.gcfs, "metcalf")) can have 3 different relationship types:

• GCF -> MolecularFamily
• GCF -> Spectrum
• Spectrum -> GCF

Questions:

1. Why do we need all three types? Can't we just use GCF -> Spectrum and derive MolecularFamily relationships through Spectrum objects when needed?

2. Analysis of the data shows:

   • Very few GCF IDs overlap between MF and Spectrum links
   • GCF IDs differ between GCF -> Spectrum and Spectrum -> GCF links (I was expecting these to match)
     What am I missing here?

For now, I'm only using GCF -> Spectrum links for the genomics -> metabolomics tab. Does this make sense?

@gcroci2 regarding the different version of the Metcalf correlation score, this was described here, briefly: the standardised score for each prospective link takes into account the sizes of the GCF and the MF and adjusts the score accordingly, making the scores comparable between links involving strain sets of different sizes as is necessary when, for example, comparing scores for different spectra or MF for a particular GCF.

Regarding the different relationship types, I think it makes sense to keep both spectrum and MolecularFamily in, as at the level of MolecularFamily, the strain membership is different than at the level of Spectrum - akin to GCF versus BGC. Thinking of it now, we do not treat the BGCs in the same way as spectra, something we could consider in the future. Happy to explain further in person, as this can get confusing quite quickly....