blog post for "on the fly transformations"

[orbeckst]

How about a blog post highlighting what @jbarnoud called "pure magic":
MDAnalysis/mdanalysis#1987 (review)

I needed it today, and it works just as expected. Along side @davidercruz's on-the-fly transformation it is pure magic.

(Perhaps also throw in notes on the new smart ChainReader by @kain88-de , especially for the Gromacs folks.)

[richardjgowers]

So @davidercruz 's transformations are going to be pretty cool for analysis. WRT unwrap, I started work on tightly integrating this with AtomGroup, so hopefully soonish we'll have ag.center_of_mass(unwrap=True) to calculate a "True" center of mass, rather than the pbc mess we currently have

[richardjgowers]

So yeah I might wait on the nice integration with AtomGroup before I go shouting about unwrap too much, I think it's still a little hard to use right now

[orbeckst]

Ok – just thought I put down a marker for the not too distant future.

We also need to be clear somewhere in the docs what we mean by "unwrap" and "pbc" and putting into boxes etc – perhaps if we first do a blog post about it with a few instructive figures and then we can recycle for the docs.

Something that explains to beginners what the problems are (essentially, the Gromacs options gmx trjconv -pbc XXX -mol YYY -ur ZZZ...)

1. we do simulations under PBC,
2. how do we get "appropriate" distances (minimum image),
3. how do we get appropriate per-molecule properties (unwrapping)
4. how to get diffusion properties (unwrapping – can we do it properly for solvent particles?)
5. keeping molecules together (e.g. for visualization)
6. packing all molecules in unit cell representations (can we do this like -ur compact – I don't think so?)

Am I missing something?

cc @kain88-de and @jbarnoud

[kain88-de]

I stand by my point that this topic is good content for the next blog post that @davidercruz has to write according to the NumFOCUS rules anyway.

[orbeckst]

Someone ought to start writing something somewhere :-).

If @davidercruz starts with his blog post that includes what he has then this can be taken as a basis for an MDAnalysis post and a write-up in the docs.

[davidercruz]

@kain88-de the blog post that I did for my first evaluation already includes some details on the transformations API itself. I can make another one later that includes the transformations that have been added since.

[kain88-de]

We need more then just a list of what has been added. Currently, only a handful of people know how cool this feature is to the rest of the world we need to show it. This means having a nice little story with examples. As I have said earlier replacing a trnconv post processing script with your transformations would be a good example. It can be directly used, people know exactly what you are doing (given they work with gromacs).

[jbarnoud]

I would like @davidercruz to start this, and to start it very soon. As I see it, it should read as a user oriented demo on a real looking case. Something like centering a transmembrane protein would be nice. Ideally, the blog post should compare the process with trjconv and contain pictures. Movies would be neat too. I strongly recommend you design the content in a jupyter notebook to make sure the example work. Then the notebook could be added to the notebook repository.

[richardjgowers]

I'm pasting this from the 0.19.0 blogpost so we can use it in the 0.20.0 release post

On-the-fly trajectory transformations

GSoC student David Cruz (@davidercruz) implemented on-the-fly trajectory transformations.
Think of it as specifying additional transformations such as centering, removing periodic-boundary
artifacts, or RMSD-fitting that are applied to a trajectory before the coordinates are passed
to user code. Transformations can be stacked and associated with a trajectory. For example:

1. center on protein
2. make all molecules whole that were broken across the periodic boundaries
3. superimpose each frame to a reference structure by minimizing the RMSD ("RMSD fitting")

No intermediate trajectories have to be written, the transformations are simply applied to the
current in-memory coordinates. Although this approach incurs a computational overhead, it is generally
still much more efficient than writing out intermediate trajectories to disk because the cost
for I/O tends to be much larger than for the computation of the transformation.
This works particularly well for interactive analysis in a Jupyter notebook as well as
[visualization of trajectories with nglview][].
Furthermore, one can implement custom transformations and stack them arbitrarily.

Davide wrote an in-depth post about his [On-the-fly transformations] where he shows more examples.

[orbeckst]

Note that I (finally – sorry @davidercruz !) merged @davidercruz 's PR MDAnalysis/binder-notebook#10 with the example notebook. I hope it still works...

EDIT: Well, data files are missing MDAnalysis/binder-notebook#13