Some notes from chat with Erick today:

• There are kappa light chains and lambda light chains. Sometimes igl means lambda and sometimes light. igk always means kappa.

• let's use ^ as joining symbol on the AA side.

• let's keep the correspondence between the codons and amino acid sequences, by adding an ambiguous codon in the nucleotide side. Let's do N^N or ^^^.

• aa model has to see the wedge.

• treat wedge as an aa character, in the sense that it will get an embedding. But, then we'll be inferring selection factors of it.

• we'll have a mix of heavy chain sequences and paired sequences, so we'll want to add the wedge character at the end of all the heavy chain sequences too.

• Think about designing things so we're not locked into having just one additional AA.

• framework.load_pcp_df can read csv headers to see if there are heavy and light chains, join as we've talked about in a new function so that there are only parent and child sequences, then do what we're already doing.

• modify everything downstream, especially stuff in sequences.py, such as index-encoding

• we'll just train the embedding for this new character for now, even though that seems like a waste.

p-IgGen trains with normal and reversed sequences 🙃