MDAnalysis as universal topology reader

[PardhavMaradani]

Hi @BradyAJohnston , @yuxuanzhuang , just wanted to make a note of this before I forget.

Currently, MN is using MDAnalysis only for Trajectories - regular ones and oxdna (with a custom topology parser). MDAnalysis supports loading a single topology file as well and has support for a wide variety of topology formats. The trajectory file is optional for creating a universe. MN UI (operator based import) currently requires a trajectory file to be specified. API doesn't have this limitation as a loaded universe is passed. With this, even regular Molecule entity formats can be loaded. These would be single frame universes, but allow using the annotations we have for it, framing based on the selection phrases or AtomGroups, etc. There isn't currently support for some formats like .cif, .bcif etc, but that doesn't seem difficult to add as it is pretty extensible.

The selections for the Molecule entity seem a bit rigid and hardcoded today (one reason why I didn't venture to add more annotations there). Maybe having a generic molecular entity using MDAnalysis's universal reader will make things a lot simpler (both for an external API as well as internal implementations) and consistent across the major entities? We'd also immediately get support for a much wider variety of topology formats (like we did with the Density entity grids by using MDAnalysis GridDataFormats). Maybe something to think about for the future. Thanks

[BradyAJohnston]

I agree that unifying the back-end to be just MDAnalysis and dropping biotite would simplify and improve a lot of the code and UX.

Main reason I haven't currently headed in that direction:

• MDA doesn't read PDBx/mmCIF or .bcif (although it is getting close for mmCIF: Implementing gemmi-based mmcif reader (with easy extension to PDB/PDBx and mmJSON) MDAnalysis/mdanalysis#4712)
• MDA doesn't currently have support for pulling out other PDB / CIF components such as symmetry operators for biological assembly

Bringing that functionality either directly into MDA or just into MN would allow us to unify everything. Molecule and Entity could be unified and it would reduce a large amount of our code required, and result in less confusion for an API user why selection strings could be used on some entities but not others.

An intermediate step could be parsing files with biotite and then having a AtomArray -> mda.Universe converter (which would also be a useful thing to have separate to MN). That would however increase overhead to open files, but no idea how much.

[PardhavMaradani]

Hi Brady, a custom MDAnalysis converter that converts from Biotite's AtomArray or AtomArrayStack to a mda.Universe might be a good intermediate step as you suggested. I'll give it a quick try over the weekend to see if there are any issues and collect some more data. Thanks

[BradyAJohnston]

Great! Looking forward to what you find. A converter from biotite structures to MDA would be welcome into the main MDA library as well!

[PardhavMaradani]

I created a converter as part of PR #960 and added more details about what I found out in the process. Thanks