Switching Scoring Methods

[shendricks464]

Good day

Metcalf did not work for me and I am trying to use the other scoring methods. However I am having difficulties enabling NPClass and Rosetta. I am able to see it and select it but when I run it it always the output "invalid scoring method". I am using NPLinker 2.0.0b1 and python 3.11.

---

ValueError Traceback (most recent call last)
Cell In[8], line 16
13 print(npl.strains)
15 # compute the links for the first 3 GCFs using rosetta scoring method
---> 16 link_graph = npl.get_links(npl.gcfs[:50], "rosetta")
18 # get links as a list of tuples
19 link_graph.links

File ~/miniconda3/envs/npl-3.11/lib/python3.11/site-packages/nplinker/nplinker.py:257, in NPLinker.get_links(self, objects, scoring_method, **scoring_params)
255 # Validate scoring method
256 if scoring_method not in self._valid_scoring_methods:
--> 257 raise ValueError(f"Invalid scoring method {scoring_method}.")
259 # Check if the scoring method has been set up
260 if not self._scoring_methods_setup_done[scoring_method]:

ValueError: Invalid scoring method rosetta.

[CunliangGeng]

Thank you for using NPLinker. The 2.0.0b1 supports only Metcalf scoring method at the moment, that's why other scoring methods are invalid.

But why doesn't Metcalf work for you? Could you please provide more detail?

[shendricks464]

Thank you for the response. Metcalf seems to be generating a lot of false links which is why I wanted to try a different scoring method.

1. If I were to use an older version of NPLinker to use a different scoring method would it be able to support GNPS2 and AntiSMASH 8.0? Also how different would the data set up be?

2. How many GCFs do you recommend for parsing for metcalf scoring?

3. For 2.0.0b1 I am only allowed to select RAW. How do I select the standardized option?

[CunliangGeng]

@shendricks464

If I were to use an older version of NPLinker to use a different scoring method would it be able to support GNPS2 and AntiSMASH 8.0? Also how different would the data set up be?

No, nplinker v1.x does not support GNPS2. The format of GNPS2 output changed a lot. You could have a look at some examples of GNPS1 and GNPS2 data in this folder.

How many GCFs do you recommend for parsing for metcalf scoring?

@justinjjvanderhooft Do you have any recommendation?

For 2.0.0b1 I am only allowed to select RAW. How do I select the standardized option?

You can do that by providing some parameters to the get_links method:

# raw score
link_graph = npl.get_links(npl.gcfs, "metcalf") 

# standardized score
link_graph = npl.get_links(npl.gcfs, "metcalf", cutoff=0.5, standardised=True)

[justinjjvanderhooft]

As for 2. - as the Metcalf (and the standardized version) are correlation-based scores, the more GCFs are involved, the better; but, more importantly, the more strains, the better, and with overlapping and contrasting GCFs to have sufficient power to pull out potentially relevant links between GCFs/BGCs and MFs/mass spectra.