Following @alexrockhill addition of https://mne.tools/stable/auto_examples/preprocessing/eeg_bridging.html and #10587

What about the case where more than 2 electrodes are bridged together, e.g. where a trio of electrodes are bridged together?

interpolate_bridged_electrodes only works on bridge pair by pair. But if we take the example below, we can do better for the bridge ('P5', 'PO7', 'PO3', 'O1').

I propose 2 changes to interpolate_bridged_electrodes:

• handles a group of electrodes bridged together as a single bridge and computes a single virtual channel per group at the average location.
• add a warning if you use interpolate_bridged_electrodes where you have bridges including more than N electrodes.

I think a good number would be N=3, a trio of electrodes is usually either a straight line on a row/column of the cap, or a triangle. If it's a row/column (e.g. PO8, PO6, PO4) the average location is actually the center electrode (PO6). If it's a triangle, the average location is more or less equidistant from the 3 electrodes. In both cases, intuitively, I think the virtual channel will improve interpolation. Above 3 electrodes, I'm not convinced.

To retrieve the group of electrodes, I can think of an implementation as a graph with networkx, something along the lines of:

G = nx.Graph()
for bridge in bridged_idx:
    G.add_edge(*bridge)
groups_idx = nx.connected_components(G)

Any other idea to avoid the dependency on networkx?