Enigmatic networkx error: " NetworkXPointlessConcept: Connectivity is undefined for the null graph."

[apayne97]

Here's my cinnabar csv:
cinnabar_input.csv

I don't know what to make of this error but here's my code:

from cinnabar import femap
fe = femap.FEMap.from_csv('cinnabar_input.csv')
fe.generate_absolute_values()

I was following the instructions here:
https://docs.openfree.energy/en/latest/tutorials/plotting_with_cinnabar.html

Environment

# Name                     Version          Build            Channel
openfe                     1.8.0            pyhd8ed1ab_0     conda-forge
openfe-analysis            0.3.1            pyhd8ed1ab_0     conda-forge
openfe-base                1.8.0            pyhd8ed1ab_0     conda-forge
cinnabar                   0.5.0            pyhd8ed1ab_0     conda-forge

Traceback:

---------------------------------------------------------------------------
NetworkXPointlessConcept                  Traceback (most recent call last)
Cell In[20], line 2
      1 fe = femap.FEMap.from_csv('cinnabar_input.csv')
----> 2 fe.generate_absolute_values()

File ~/miniconda3/envs/openfe/lib/python3.13/site-packages/cinnabar/femap.py:456, in FEMap.generate_absolute_values(self)
    453 if not all(d['DG'].u == u for _, _, d in mes):
    454     raise ValueError("All units must be the same")
--> 456 if self.check_weakly_connected():
    457     graph = self.to_legacy_graph()
    458     f_i_calc, C_calc = stats.mle(graph, factor="calc_DDG")

File ~/miniconda3/envs/openfe/lib/python3.13/site-packages/cinnabar/femap.py:442, in FEMap.check_weakly_connected(self)
    439         continue
    440     comp_graph.add_edge(a, b)
--> 442 return nx.is_connected(comp_graph)

File ~/miniconda3/envs/openfe/lib/python3.13/site-packages/networkx/utils/decorators.py:784, in argmap.__call__.<locals>.func(_argmap__wrapper, *args, **kwargs)
    783 def func(*args, __wrapper=None, **kwargs):
--> 784     return argmap._lazy_compile(__wrapper)(*args, **kwargs)

File <class 'networkx.utils.decorators.argmap'> compilation 13:4, in argmap_is_connected_9(G, backend, **backend_kwargs)
      2 import collections
      3 import gzip
----> 4 import inspect
      5 import itertools
      6 import re

File ~/miniconda3/envs/openfe/lib/python3.13/site-packages/networkx/utils/backends.py:551, in _dispatchable._call_if_no_backends_installed(self, backend, *args, **kwargs)
    545 if "networkx" not in self.backends:
    546     raise NotImplementedError(
    547         f"'{self.name}' is not implemented by 'networkx' backend. "
    548         "This function is included in NetworkX as an API to dispatch to "
    549         "other backends."
    550     )
--> 551 return self.orig_func(*args, **kwargs)

File ~/miniconda3/envs/openfe/lib/python3.13/site-packages/networkx/algorithms/components/connected.py:205, in is_connected(G)
    203 n = len(G)
    204 if n == 0:
--> 205     raise nx.NetworkXPointlessConcept(
    206         "Connectivity is undefined for the null graph."
    207     )
    208 return len(next(connected_components(G))) == n

NetworkXPointlessConcept: Connectivity is undefined for the null graph.

[jthorton]

Hey @apayne97 thanks for the issue and the csv file that helped a lot! I hit a few issues getting this input to work.

First the error is advising that no edges are in the graph constructed from the csv file, this is due to some poorly documented specific requirements of the input format - it needs 5 columns for the calculated section the two ligand labels the calculated value, mbar and other error. I added an extra column to your input full of zeros which made the expected graph!

Next I could not calculate the absolute values with MLE due to their being some edges with an mbar uncertainty of exactaly zero in your input which causes the solver to fail I have opened a PR which pad these results with a tinny value automatically in future.

Lastly I noticed that some of the uncertainties are negative when they should all be positive, converting them to postivie allowed me to generate the plots.

Here is the updatted csv file, you will need to use the branch in this PR to generate the absolute values or pad the zero uncertainties in the csv file.

cinnabar_input.csv

Finally since the csv format is full of issues you might want to consider using the API to do this see this tutorial for more details.

Let me know if you have any more issues!

[apayne97]

@jthorton wow nice sleuthing. Thanks for the help! I guess some of these requirements should be obvious but it would be nice if there was a dataclass for the cinnabar input that would validate the input and give helpful error messages....maybe identifying the most likely errors is kinda hard.

I made the csv based on https://docs.openfree.energy/en/latest/tutorials/plotting_with_cinnabar.html.
I ignored the "calc_dDDG(additional)" column because it didn't write anything to that column - didn't occur to me that it would need to be populated with zeros!

As you may have noticed this is data for a radial network. Do you think that is causing issues with the MLE solver?

[jthorton]

I guess some of these requirements should be obvious but it would be nice if there was a dataclass for the cinnabar input that would validate the input

Yeah thats a great point for now only the API ensures this kind of validation but maybe some more errors might be good for the csv parsing, however I could see a future where we phase this out as the input format is just ugly and the API is simpler to use.

As you may have noticed this is data for a radial network. Do you think that is causing issues with the MLE solver?

This is an interesting one, due to the use of a purely radial network, the application of MLE should make no difference to the predicted DDGs backcalculated from the MLE DGs so your results depend purely on the simulated DDGs. You will need some extra legs in the network to impose additional constraints on the differences between non-central ligands, to potentialy improve the predictions. See the docs on MLE for more detailed explanation on how it works.

The other thing to consider is the large uncertainties on some of your calculated values I would look into these and check for mistakes with some of them also being negative.