MMseq2 API error

Hi,

I ran into an issue when using the local colabfold API. I tried to produce models for a heterodimer sequence but I encounter a 'MMseq2 API error'.

Input:
colabfold_batch queries predictions --host-url http://localhost --amber --recycle-early-stop-tolerance 0.1 --num-recycle 20 --num-models 3 --rank plddt --use-gpu-relax

Input sequences in MULTI-fasta format (one example):

>heterodimer_0-100
RTQEKEQIKTLNNKFASFIDKVRFLEQQNKMLETKWSLLQQQKTARSNMDNMFESYINNLRRQLETLGQEKLKLEAELGNMQGLVEDFKNKYEDEINKRT:
IQNEKETMQSLNDRLASYLDRVRSLETENRRLESKIREHLEKKGPQVRDWSHYFKIIEDLRAQIFANTVDNARIVLQIDNARLAADDFRVKYETELAMRQ

Error:

2024-09-05 12:01:00,416 Could not get MSA/templates for heterodimer_0-100: MMseqs2 API is giving errors. Please confirm your input is a valid protein sequence. If error persists, please try again an hour later.
Traceback (most recent call last):
  File "/usr/local/xray/conda/colabfold/colabfold-conda/lib/python3.10/site-packages/colabfold/batch.py", line 1468, in run
    = get_msa_and_templates(jobname, query_sequence, a3m_lines, result_dir, msa_mode, use_templates,
  File "/usr/local/xray/conda/colabfold/colabfold-conda/lib/python3.10/site-packages/colabfold/batch.py", line 844, in get_msa_and_templates
    paired_a3m_lines = run_mmseqs2(
  File "/usr/local/xray/conda/colabfold/colabfold-conda/lib/python3.10/site-packages/colabfold/colabfold.py", line 239, in run_mmseqs2
    raise Exception(f'MMseqs2 API is giving errors. Please confirm your input is a valid protein sequence. If error persists, please try again an hour later.')
Exception: MMseqs2 API is giving errors. Please confirm your input is a valid protein sequence. If error persists, please try again an hour later.
2024-09-05 12:01:00,422 Done

I looked into the .a3m files which are produced and found that sequence 1 is correctly indicated as '>101' but sequence 2 is incorrectly indicated as '\00>102'. I made some changes into the batch.py and other .py so I would like to know if this is due to me screwing up the .py or if this is a bug.

Kind regards,
Gistal.

Start by reproducing the provided colabfold_batch command and inspect how the MULTI-fasta input is converted into .a3m headers. Read the relevant paths in batch.py and colabfold.py, then compare the generated headers with the MMseqs2 request and response. Done means determining whether the malformed '\00>102' header is produced by ColabFold or by the local changes and documenting a reproducible fix or cause.