as mentioned in some emails a very long time back, my colleagues are into immune repertoire sequencing with Illumina machines. They have reached the state now that they have quite some sequences available and do not really now what to do with them - beyond collecting and counting. My particular ambition always was to find equivalence classes (my mathematically motivated term for functional clusters) of their paratopes. The ways towards such should be via antibody.py, obviously.
The sequences provided are the C-terminal bits of the heavy chain - no light chain is looked at. The reasoning is that past experiments (as in 20 years past, see http://www.jimmunol.org/content/155/6/3223.long) had shown for a mouse model of Lupus, and our phage-display experiments are mostly suggesting the same, that the pathogenicity was kept up for most of the many substitutes of the light chain, indicating the heavy chain as the prime driver of the phenotype. It will take another while to get single-cell PCR up and running to investigate pairs, so there is no real alternative to single chains for us, anyway. A consequence for me is that a predictor of antibody structures should also only see the heavy chain and individually pick a reasonable light chain for it. "Reasonable" here as in "not ultimately likely to stress the the heavy chain".
The patch presented in this pull request - it is more of a request for discussion at the very moment - performs a blast search with the heavy chain in the antibody_database and the light chain of the structure identified as the most similar is substituting what the user would otherwise have submitted. The script then continues as ever. The patch looks a bit rough as I induced a big rip to two functions, separating the treatment of the light and heavy chains into functions of their own. The intellectual challenge is overseeable, I just hope not to have messed it up, anyway. Local testing was promising but extra eyeballs are appreciated.
Questions:
Do you like this "heavy-chain-only" addition to antibody.py in principle? Do you possibly even know someone who would like to use it?
The similarity search with the whole heavy chain is not unreasonable, but there is little reason to involve the more variable loops in the search - should the search possibly involve the more constant inter-loop regions more? Are there other ideas? How much is missing for an application's note to get the news into the world?
To be useful for the NGS folks with their fragmental Fv regions, a web service would first need to consult e.g. the IMGT V-Quest server to reasonably complete what is almost there. Would there be a place for this on the ROSIE? After all, the submission is likely to be on a nucleotide level.
lyskov
commented
6 years ago
Dear all,
as mentioned in some emails a very long time back, my colleagues are into immune repertoire sequencing with Illumina machines. They have reached the state now that they have quite some sequences available and do not really now what to do with them - beyond collecting and counting. My particular ambition always was to find equivalence classes (my mathematically motivated term for functional clusters) of their paratopes. The ways towards such should be via antibody.py, obviously.
The sequences provided are the C-terminal bits of the heavy chain - no light chain is looked at. The reasoning is that past experiments (as in 20 years past, see http://www.jimmunol.org/content/155/6/3223.long) had shown for a mouse model of Lupus, and our phage-display experiments are mostly suggesting the same, that the pathogenicity was kept up for most of the many substitutes of the light chain, indicating the heavy chain as the prime driver of the phenotype. It will take another while to get single-cell PCR up and running to investigate pairs, so there is no real alternative to single chains for us, anyway. A consequence for me is that a predictor of antibody structures should also only see the heavy chain and individually pick a reasonable light chain for it. "Reasonable" here as in "not ultimately likely to stress the the heavy chain".
The patch presented in this pull request - it is more of a request for discussion at the very moment - performs a blast search with the heavy chain in the antibody_database and the light chain of the structure identified as the most similar is substituting what the user would otherwise have submitted. The script then continues as ever. The patch looks a bit rough as I induced a big rip to two functions, separating the treatment of the light and heavy chains into functions of their own. The intellectual challenge is overseeable, I just hope not to have messed it up, anyway. Local testing was promising but extra eyeballs are appreciated.
Questions:
Best regards,
Steffen