DSSP guess_hydrogens=True produces small secondary-structure mismatches vs reference DSSP outputs

[sushk2904]

Expected behavior

The test test_file_guess_hydrogens in MDAnalysisTests.analysis.test_dssp is expected to pass when DSSP is run with guess_hydrogens=True, producing secondary structure assignments that match the reference .dssp files shipped with MDAnalysisTests.

Actual behavior

When running the DSSP hydrogen-guessing tests locally, multiple failures occur in test_file_guess_hydrogens.
Specifically:

• Out of 100 selected tests, 22 fail and 78 pass
• Failures occur consistently for the same PDB files
• The failures appear both with the multiprocessing backend and the serial backend
• The mismatches are small and localized, typically involving helix (H) or strand (E) boundary assignments, often differing by a single residue
• The output strings always have the same length and overall structure, but exact string comparison fails due to these small boundary shifts.

This suggests an environment or execution-sensitive behavior rather than an installation or setup issue.

Code to reproduce the behavior

The issue can be reproduced by running only the DSSP hydrogen-guessing tests:

pytest --pyargs MDAnalysisTests.analysis.test_dssp -k guess_hydrogens -vv

This selects 100 tests (12 deselected), of which 22 fail on my system.
Failures occur for the same structures under both client_DSSP0 (multiprocessing) and client_DSSP1 (serial).
Example failing structures include (3gknA.pdb.gz, 2j49A.pdb.gz, 3aqgA.pdb.gz, 3nzmA.pdb.gz, 3hklA.pdb.gz, 3l4rA.pdb.gz, 3e8mA.pdb.gz, 2fvvA.pdb.gz, 3a4rA.pdb.gz, 3gfsA.pdb.gz, 1eteA.pdb.gz etc.)

All failures show similar off-by-one or boundary-level differences in DSSP assignments.

Current version of MDAnalysis

• MDAnalysis version: 2.11.0-dev0 (editable install from source)
• Python version: 3.11.0rc1
• Operating system: Ubuntu 22.04 (Linux)

Additional context:

• MDAnalysis built from source with C extensions compiled
• MDAnalysisTests installed via pip
• The broader test suite passes, with failures isolated to DSSP hydrogen guessing
  @orbeckst

[rishitha2827]

Hi! I’d like to work on this issue and submit a PR.

I can reproduce the failures locally and will investigate whether the
boundary-level DSSP differences with guess_hydrogens=True should be
handled by relaxing the test comparison or by adjusting the DSSP wrapper logic.

Please let me know if that approach sounds reasonable.