Currently this repo is missing scripts to recalculate embeddings.npz. This step involves pulling structures and summary data from Sabdab, pulling the sequences from IMGT-numbered residues 1 to 128, clustering with MMSeqs to a 70% threshold, taking the highest-resolution structures from each cluster, and embedding them using SoftAlign. This should ideally be run as a github action, but unless sequence data can be fetched without downloading all IMGT-renumbered structures, it might take too long.
Currently this repo is missing scripts to recalculate
embeddings.npz. This step involves pulling structures and summary data from Sabdab, pulling the sequences from IMGT-numbered residues 1 to 128, clustering with MMSeqs to a 70% threshold, taking the highest-resolution structures from each cluster, and embedding them using SoftAlign. This should ideally be run as a github action, but unless sequence data can be fetched without downloading all IMGT-renumbered structures, it might take too long.