Small inconsistency with PTMs in predicted peptides

[aukeheerdink]

Hi, I've been playing around with the newest release for a bit and it seems to work great! Really nice work. While playing around and analysing the peptides that were predicted, I noticed a small inconsistency with the N-terminal modifications. I don't know if you are following the ProForma notation but the output seems really similar so I assumed so. For as far as I could tell InstaNovo is compliant with the specification except for this small part. For example this peptide: [UNIMOD:5]GEEEEDDDSLSK (currently in my dataset) should be [UNIMOD:5]-GEEEEDDDSLSK (ProForma compliant) for N-terminal modifications. Would it be possible to adjust the predicted peptide so it displays the N-terminal modifications like this? This way more people and downstream tools are able to parse these peptides in a standardised way.

[BioGeek]

Hi Auke,

Thanks for bringing this to our attention.

I can confirm that our current implementation of N-terminal modificants is not ProForma compliant.

>>> from pyteomics import proforma
>>> proforma.parse('[UNIMOD:5]GEEEEDDDSLSK')
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/home/j-vangoey/code/InstaNovo/.venv/lib/python3.12/site-packages/pyteomics/proforma.py", line 2024, in parse
    raise ProFormaError(
pyteomics.proforma.ProFormaError: Pyteomics error, message: 'Error In State ParserStateEnum.post_tag_before, unexpected G found at index 11' ('Error In State ParserStateEnum.post_tag_before, unexpected G found at index 11', 11, <ParserStateEnum.post_tag_before: 10>)
>>> proforma.parse('[UNIMOD:5]-GEEEEDDDSLSK')
([('G', None), ('E', None), ('E', None), ('E', None), ('E', None), ('D', None), ('D', None), ('D', None), ('S', None), ('L', None), ('S', None), ('K', None)], {'n_term': [UnimodModification('5', None, None)], 'c_term': None, 'unlocalized_modifications': [], 'labile_modifications': [], 'fixed_modifications': [], 'intervals': [], 'isotopes': [], 'group_ids': [], 'charge_state': None})

It is important for us to get this right, because for example denovo_benchmarks also parses the predictions with ProFroma.

We'll fix this in a next release.