GridDataFormats refuses to load 3D MRC files with space group 0

[PlethoraChutney]

MDAnalysis strictly follows the CCP4 and MRC file specs which state that files with a value of zero for the ISPG header field are stacks of 2D images, not single 3D volumes. However, many cryo-EM and volume visualization packages write a zero to this field because there is no "space group unknown" value. Because of this deviation from spec, most visualization packages open space group 0 files as volumes. For more from their point of view, see this bug ticket for ChimeraX.

Downstream users of GridDataFormats may be surprised that it refuses to open volumes that successfully open elsewhere. While GridDataFormat's behavior is formally correct, I think in practice it may be better to open any MRC file with a 3D data array as a 3D volume.

[orbeckst]

Context by @BradyAJohnston #149 (comment)

It's related to this issue (BradyAJohnston/MolecularNodes#993) raised in MN. Downloading directly from the EMDB it would load fine, but if we round-trip through ChimeraX then it would fail to load with Grid() because of the points specified above.

[orbeckst]

I generally do not like automatically ignoring standards because you implicitly assume you know what the user wants. And that's impossible. It also specializes a library unnecessarily — as Tom Goddard pointed out, ChimeraX does not care about space groups, so it makes sense for it to not pretend it knows what to put into the file.

I'd rather have the user opt-in to standard violations. For example, we could add a keyword argument to the MRC file parser is_volume=None|True|False; the default None would strictly enforce the standard based on the file (how it's currently done), is_volume=True allows the user to override the check (and False is there for orthogonality). We can pass **kwargs from Grid to the file parsers and have the parsers ignore everything they do not care about. Users could then write

g = Grid("chimera.mrc", is_volume=True)

which clearly documents the user's intention to treat all mrc files as volume maps because the user knows the specific context in which the file is produced and loaded.

Would this be a possible path forward?

[PlethoraChutney]

I think this makes sense! Better to follow the spec by default!