on running colabfold locally

Hi there,

We have plans on running a few hundreds of complexes (some relatively big) and we've been trying to set up a pipeline to run locally on our HPC (actual local run not remote access to the notebooks).

I've been able to run colabfold_search.sh locally thanks to @milot-mirdita's comment on #70. I used a few complexes as input (just dimers) and the MSA generation ran until completion. One of my questions is the role of merge_and_split_msas.py, which based on the comments in the script is meant to separate the MSA's(?), well, I only got one single a3m file as output named based on the first fasta header of the input (out of 20).

Another question, colabfold_batch still uses the Google Colab notebooks, not local resources, so it's not an option. I was wondering if there is an easy way to 'pickle' the MSA files into the default intermediate msa.pickle that the advanced notebooks used to generate. It is the standard input for localcolabfold so it would makes things easy as we could run localcolabfold for the model generation part once the multimer option is released.

Thank you in advance.

Start by reading colabfold_search.sh and merge_and_split_msas.py to understand how the input complexes produce A3M files. Then inspect colabfold_batch, localcolabfold, and the advanced notebooks around msa.pickle; the work is complete when the local workflow and its expected MSA format are clearly established or supported.