[hannahbaumann]

When two chains of a protein have the same segid, the PBC are not handled correctly during alignment since they would appear as one segment. This PR switches to use the protein.fragments instead. For fragments to be recognized correctly, we first need to guess the bonds in the protein.
Having the large fragments now makes unwrap() very slow which is fragment based.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 98.38%. Comparing base (988b7de) to head (933e835).

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #124      +/-   ##
==========================================
+ Coverage   98.35%   98.38%   +0.02%     
==========================================
  Files          10       10              
  Lines         487      496       +9     
==========================================
+ Hits          479      488       +9     
  Misses          8        8              

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[hannahbaumann]

This now makes the alignment very very slow. Unwrap loops over the fragments and before (when there was no bond information), there were a lot of fragments that were very small.

[IAlibay]

How much of a problem is this in practice?

[hannahbaumann]

@IAlibay I just checked the timing, for a single lambda windows, 51 frames, the applying of the transformations took ~16 sec, but since this happens multiple times, each time for each analysis, this is x4.
I just switched it to doing the applying the alignments first and storing it in memory, no it's faster. It will still be a considerable cost for e.g. ABFE with the many lambda windows though.

[IAlibay]

One thing to fix, one question, and a missing news entry.

Otherwise it looks great to me!

[IAlibay]

How much of a problem is this in practice?

[IAlibay]

Is it worth checking for bonds in protein before you do this?

[hannahbaumann]

Added a check above.

[hannahbaumann]

@IAlibay here some timing data on a SepTop A2A system, 100 frames.

19 states x 101 analyzed frames

stage                   calls   total (s)    % wall
build_universe             19       0.129      0.1%
register_transforms     19       4.096      1.7%
transforms                    19     217.842     90.7%
protein_2D_RMSD            19       1.257      0.5%
ligand_RMSD                38       2.483      1.0%
ligand_COM_drift           38       0.036      0.0%
other (plot/npz/io)                14.226      5.9%
WALL TOTAL                        240.06

So it's on the order of minutes, would this be ok? I'd have to update the structural analysis in SepTop and ABFE to tranfer the universe to memory to not redo the transformations for each analysis class.

[IAlibay]

this looks good to me, thanks