Update nallo rank model #117
Conversation
+3 regardless of model Add polyphen, revel, sift, dbnsfp_gerp++_rs, dbnsfp_phastcons100way_vertebrate, dbnsfp_phylop100way_vertebrate Add most severe pli for SVs update rank model suggestion
fellen31
force-pushed
the
update-nallo-rank-model
branch
from
946238e to
85fcb39
Compare
dnil
left a comment
dnil
left a comment
There was a problem hiding this comment.
Nice! Good test results overrides any guesswork we might have on scores. Note the diff on SV loqusdb not_reported/missing scores (4 vs 6): is it intentional and still valid?
| [[common]] | ||
| score = -12 | ||
| lower = 0.1 | ||
| lower = 0.02 |
| [[not_reported]] | ||
| score = 4 | ||
|
|
||
| [[missing]] |
There was a problem hiding this comment.
Right, are we still getting these? 🤔 Oh well, good fallback.
There was a problem hiding this comment.
Yeah, never got around to fixing it in echtvar..
| upper = 0.01 | ||
| upper = 0.0005 | ||
|
|
||
| [revel] |
| lower = 0.75 | ||
| upper = 1 | ||
|
|
||
| [dbnsfp_gerp++_rs] |
| lower = 0 | ||
| upper = 2 | ||
|
|
||
| [dbnsfp_phastcons100way_vertebrate] |
| lower = 0 | ||
| upper = 0.8 | ||
|
|
||
| [dbnsfp_phylop100way_vertebrate] |
| score = 4 | ||
| lower = 0 | ||
| upper = 0.01 | ||
| upper = 0.0005 |
| separators = ',', | ||
|
|
||
| [[not_reported]] | ||
| score = 4 |
| [[common]] | ||
| score = -12 | ||
| lower = 0.1 | ||
| lower = 0.05 |
| lower = -400 | ||
| upper = -1 | ||
|
|
||
| [gene_intolerance_score] |
|
|
||
| [[not_reported]] | ||
| score = 0 | ||
| score = 3 |
There was a problem hiding this comment.
This feels weird, but again, the results are most important. The genotypes for SVs are still too noisy I suppose. 😞
There was a problem hiding this comment.
We don't have any variants that doesn't get a model, so in practice this doesn't seem to matter. I can change it back next update.
There was a problem hiding this comment.
Yikes, yes, then no worries. Something to bump over at the CNV callers? We added a simple model for CNV-nator way back when, to at least get some genotypes for the copy number changes, where we have some decent statistics to work with. Is that better in Sawfish compared to HiFiCNV by any chance?
Yes, I tried giving +6 but I think that brings too many small intronic indels too high in the ranking. I don't think the difference between something that has been seen once in loqusdb (gets +2) and something completely new should be that big, I wonder if there even should be a difference in score. |
No, not really ever. The intention is that once the db grows to a size where the somewhat rare threshold is well estimated one can start differentiating. But indeed, that would likely need a few thousand cases with that EDIT: and yes, maybe not for a long time, given a lot of small intronic events. Until we have some functional predictors for them perhaps. |
Description
Added
Changed
How to prepare for test
uspaxaHow to test
Expected test outcome
Review
Thanks for filling in who performed the code review and the test!
This version is a
Implementation Plan