This looks pretty good. I added some comments for improvements that could be made.

I would also like to see the addition of a standalone pvacseq add_ml_predictions command that someone could run standalone to add ml_predictions to the output from their pipeline if the ml option was not enabled in the original run. Have a look at pvactools/tools/pvacseq/mark_genes_of_interest.py and its test tests/test_pvacseq_mark_genes_of_interest.py to see how to do so. This command addition also requires an update to pvactools/tools/pvacseq/main.py and pvactools/tools/pvacseq/__init__.py to add this command.

Additionally, this new command should then be documented in docs/pvacseq/optional_downstream_analysis_tools.rst. Most of the documentation can be auto-generated from the command line docs by using the program-output tool but this would also be the spot to put any additional documentation you want to have around this feature.

Left a couple of code suggestions to convert named required parameters to positional parameters in your parser.

+1

I made some copy editing suggestions. Two thoughts I had while reviewing this:

• It would be great to have some additional exploration of different candidates in ML section of the pVACview vignette. I wrote down some examples in a comment in that part of the text.
• Have you and Malachi discussed whether it would be worthwhile to make the reject threshold configurable? It seems a bit weird to me to hardcode the reject threshold but make the accept threshold configurable.