HLAthena Docker invalid_grant error and intermittent web server disconnection

[BBG303 xzywis]

Hi HLAthena team,

I am trying to run HLAthena predictions for peptide-format inputs with context sequences, but I have encountered issues with both the Docker version and the web interface.

My input file is tab-delimited with a header. The columns are:

n_mer    ctex_up    ctex_dn    TPM

The peptide column is named n_mer. I set ctex_up and ctex_dn to 30 amino-acid residues, padding missing upstream/downstream residues with -. I also removed non-standard amino-acid symbols. In my current tests, I do not use expression values, so the TPM column is either blank or ignored.

For the Docker version, I pulled the image successfully:

docker pull ssarkizova/hlathena-external

The help command works:

docker run --rm -v "${work}:/work" -w /work ssarkizova/hlathena-external predict --help

However, when I run prediction, the process fails after cleavability prediction. For example, this command:

docker run --rm -v "${work}:/work" -w /work ssarkizova/hlathena-external predict \  --runID test08_A0201 \  --rundir ./hla_test_A0201/ \  --peptides ./test_08mers_100_fixed.tsv \  --peptide_col_name n_mer \  --alleles A0201 \  --run_ctex true \  --expr_col_name TPM \  --logtransform_expr true \  --assign_by_ranks_or_scores ranks \  --assign_threshold 0.1 \  --aggregate_pep false

runs through the cleavability step and prints messages such as:

found ctex in header predicting cleavability... All done! found peptides of length: 8 copying panpan models

but then fails with:

invalid_grant: Invalid JWT Signature.
CommandException: 1 file/object could not be transferred.

I also reproduced the same issue with a larger allele list. It seems that the input file is read correctly and the cleavability model runs, but the workflow fails when copying or accessing the panpan models.

Separately, I also tried the web interface at http://hlathena.tools/. The sample peptide input with A0101 worked successfully and produced downloadable results. My own fixed 100-peptide test file with A0101 also worked. A batch of about 2,000 peptides and 5 alleles also completed once, taking around 4 minutes.

However, the web interface later became unstable. After uploading a file and entering alleles, clicking “Predict” sometimes immediately gives:

Error: An error has occurred. Check your logs or contact the app author for clarification.

After refreshing the page, it shows:

Disconnected from the server. Reload

Clicking Reload still leaves the page disconnected.

Could you please advise on the following?

1. Is the Docker image currently expected to work, or is the invalid_grant: Invalid JWT Signature error related to an expired or inaccessible Google Cloud/model credential inside the image?

2. Is there a recommended workaround for running the Docker version locally without this panpan model transfer error?

3. Is the web server currently experiencing instability or resource limits?

4. For web prediction, what is the recommended maximum number of peptides and alleles per job when using context/MSiC prediction?

Thank you very much for your help.