Open iskandr opened 8 years ago
+1
Aside from not needing to run MHC binding prediction over the proteome, remind me how that differs in output? Is it the same unless we were to do approximate matching with the proteome, in which case we'd want to run MHC binding prediction for additional peptides (possibly via the self-ligandome data)?
I think all we'd be giving up is the ability to later use inexact string matches (such as looking at only interior TCR-facing residues).
Given that we aren't doing that currently, 👍
Sacha Gnjatic's group is incorporating minimal Levenshtein distance between a predicted neoepitope and all members of the self-ligandome as a metric for peptide ranking. I agree that filtering against the self-proteome is a good (perhaps best) default behavior but I wouldn't dismantle the self-ligandome components just yet unless they are impeding progress.
I'm not sure it's worth the extra complexity to filter candidate epitopes by a "self-ligandome" set of peptides. No one other than Kipp ever generated these peptide sets and no one seems to use them. So, in the interest of using a more precise filter, Topiary ended up without any kind of self-filtering whatsoever. Simpler alternative: chop up the reference proteome from pyensembl and check for membership in that peptide set.