[Biopython] Overhauling of Bio.PDB module

[Joe Greener]

Hi Patrick,

Some of us spoke about this at CoFest too, inspired by the ideas in Biotite (I don't think you and I spoke at BOSC though). As I recall it was João, Spencer, myself and possibly Peter in the discussions.

We were in favour of the fundamental idea of a large coordinate array that is indexed into. As you point out though it would be no small amount of work to implement. I personally won't have time to do it, though I am happy to discuss and review code.

I view Bio.PDB like a beloved older car that has been patched up over many years. It is probably the most widely used and debugged PDB parsing code around, and any overhaul would have to make sure to maintain the behaviour that many people rely on. That said, it does have its peculiarities and is rather slow (https://github.com/jgreener64/pdb-benchmarks). I'm just saying that we should make sure to get consensus before merging any overhaul PRs. But for sure I am in favour of someone making those PRs.

Best,
Joe

Joe Greener
Research Associate, UCL
http://jgreener64.github.io


On 16/10/2019 12:37, Patrick Kunzmann wrote:
Hello Biopythoneers,

at the BOSC this year we talked about overhauling the Bio.PDB module. The problem is that currently the atom coordinates are stored in a separate NumPy array for each atom. This design prevents efficient computation of all kinds of analyses (distances, angles, superimpositions, etc.). One proposed possible solution to this problem, we talked about, was to put the coordinates of the entire structure in one NumPy array, and let the Atom, Residue, Chain and Structure objects point to positions in this array. The benefit of this approach is that functions could be directly applied onto the entire array, harnessing the power of vectorization.

For the analysis we could adapt the vectorized functions from the Python package Biotite, a project I am currently working on (https://www.biotite-python.org/apidoc/biotite.structure.html). Usually, these functions already accept the coordinates as NumPy array, so I think only a few tweaks would be necessary for every function.

However, we would require one person or a small team who makes the effort to implement the new structure types and adapts the analysis functions. I could offer a pair of helping hands in the adaption of the analysis functions, but I don't have the time for anything more.

So the question is: Is there anyone out there, who is willing to do this work? Alternatively, I would propose to write a 'bridge' package between Biopython and Biotite, that converts the Biopython structure representation into the representation in Biotite and vice versa. I think, this solution is less elegant but would also require less effort.

Best regards

Patrick Kunzmann

_______________________________________________
Biopython mailing list  -  Biopython at mailman.open-bio.org<mailto:Biopython at mailman.open-bio.org>
https://mailman.open-bio.org/mailman/listinfo/biopython
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.open-bio.org/pipermail/biopython/attachments/20191016/0c7b214c/attachment.htm>