Faulty modifications handled

[lliligabriel]

Describe the bug
Version: 1.4.0

Reproduce the bug
To Reproduce

1. peptdeep predict
2. My input file is a peptdie table that looks like

    TGNSSLCPI,Carbamidomethyl@C,7
TCKLLGINM,Carbamidomethyl@C,2
YKIGYICS,Carbamidomethyl@C,7

for example
3. run peptdeep predict with default parameters and

library:
  infile_type: peptide_table
  infile_type_choices:
  - fasta
  - sequence_table
  - peptide_table
  - precursor_table
  - all_other_psm_reader_types
  infiles:
  - peptide_table.csv
  fasta:
    protease: ([KR](?=[^P]))
    protease_choices:
    - trypsin
    - ([KR])
    - trypsin_not_P
    - ([KR](?=[^P]))
    - lys-c
    - K
    - lys-n
    - \w(?=K)
    - chymotrypsin
    - asp-n
    - glu-c
    max_miss_cleave: 5
    add_contaminants: false
  fix_mods: null
  var_mods:
  - Acetyl@Protein_N-term
  - Oxidation@M
  special_mods: []
  special_mods_cannot_modify_pep_n_term: false
  special_mods_cannot_modify_pep_c_term: false
  labeling_channels: {}
  min_var_mod_num: 0
  max_var_mod_num: 4
  min_special_mod_num: 0
  max_special_mod_num: 1
  min_precursor_charge: 1
  max_precursor_charge: 4
  min_peptide_len: 6
  max_peptide_len: 15
  min_precursor_mz: 200.0
  max_precursor_mz: 2000.0
  decoy: pseudo_reverse
  decoy_choices:
  - None
  - pseudo_reverse
  - diann
  max_frag_charge: 2
  frag_types:
  - b
  - y

Expected behavior
An accurate mapping of modifications generated. Oxidation only at M, Carbamidomethyl modification only at C, etc.

** However**

Only Carbamidomethyl modifications are generated, and at residues not only limited to C

Version (please complete the following information):
Platform information:
system - Linux
release - 5.15.0-173-generic
version - #183-Ubuntu SMP Fri Mar 6 13:29:34 UTC 2026
machine - x86_64
processor - x86_64
cpu count - 58
ram - 1063.4/1279.7 Gb (available/total)

Additional context

[lliligabriel]

Two things:

1. Regarding the random C modification - I think the issue may lie in the pseudo_reverse decoy generation:

def _decoy(self, sequence: str) -> str:
    return sequence[:-1][::-1] + sequence[-1]   

SpecLibDecoy.__init__ populates the decoy dataframe via copy.deepcopy(target_lib.__dict__)

This copies mods and mod_sites columns verbatim from the target. The reversed sequence gets the original positional indices applied to it, which now point to completely different (non-C) residues.

2. Regarding the fact that no flavors of variable modifications showed up

PeptideLibraryMaker never calls add_modifications() for inputs peptide_table and precursor_table.

Happy to submit a pull request if needed