You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Output from joint genotyping is a nice squared off matrix of variant rows and callset (sample) columns.
It is typically the case that around 90% of such data is high quality calls of homozygous reference.
This additional 90% of data is expensive to insert, store, and query, and often fundamentally unused.
If one were to only insert only the variant calls and no-calls, and you know your list of samples, you can infer for each call which samples were homozygous reference.
On one test dataset, importing a subset of total variants, we observed that dropping homozygous reference calls reduced import cost around 75%. Further it reduced the storage size (and hence cost) about 90%. The reduced storage size also means similar reduction in query costs. Other datasets will vary in how they are impacted.
I propose a new flag:
--hom-ref-disposition=['keep', 'drop']
with the default today being keep. This could be expanded in the future to allow for keeping the homozygous reference calls, but perhaps only the name (not the whole call record) or a few key fields [name, GQ, AD] in a separate hom_ref column.
The text was updated successfully, but these errors were encountered:
Output from joint genotyping is a nice squared off matrix of variant rows and callset (sample) columns.
It is typically the case that around 90% of such data is high quality calls of homozygous reference.
This additional 90% of data is expensive to insert, store, and query, and often fundamentally unused.
If one were to only insert only the variant calls and no-calls, and you know your list of samples, you can infer for each call which samples were homozygous reference.
On one test dataset, importing a subset of total variants, we observed that dropping homozygous reference calls reduced import cost around 75%. Further it reduced the storage size (and hence cost) about 90%. The reduced storage size also means similar reduction in query costs. Other datasets will vary in how they are impacted.
I propose a new flag:
with the default today being
keep
. This could be expanded in the future to allow for keeping the homozygous reference calls, but perhaps only thename
(not the whole call record) or a few key fields [name
,GQ
,AD
] in a separatehom_ref
column.The text was updated successfully, but these errors were encountered: