Question: Reuse MSA even when sequence changes?

I would like to keep the MSA for a number of predictions of different sequences. The sequences have a small number of homologous mutations (sometimes insertions or deletions), so should have a very similar MSA anyway.

When I use the `--use_precomputed_msa` flag but change the sequence in the input fasta file, I get a shape error. I assume re. incorporating the pair bias into the MSA representation. Can anyone think of a workaround?

Thanks for insight.

EDIT: I believe this occurs in the `make_msa_features` component of `data/pipeline.py`. The number of residues is set as the first sequence in the MSA. If this is true, then hypotheically replacing the first sequence in the MSA with the new one would fix it. This would require rerunning alignment of the new seq. against the existing ones, but would not require any jackhmmering? This seems like a useful feature for studies of small sets of mutations and would like to figure it out if anyone has a better idea about how to hack the existing MSA on file.