walaj
commented
7 years ago Hi Brad, Great news about wrapping into bcbio. This would definitely make this tool more accessible as part of larger pipelines.
I think your suggestion makes a lot of sense. I could add an option so that rather than discarding reads for any variant filter (coverage or otherwise), an alignment tag could be added. Another option would be to just flag them as QC fail, which would probably do the trick for most variant callers out of the box (rather than have them try and interpret a new alignment tag).
I'll give this a try over the weekend (with either flag or tag as an option). Shouldn't be more than a few lines of code to add this feature.
On Wed, Jun 7, 2017 at 6:15 AM, Brad Chapman notifications@github.com wrote:
Jeremiah; Thank you for VariantBam, it's a really useful tool and we're looking to incorporate it's maximum coverage filtering into bcbio ( https://github.com/chapmanb/bcbio-nextgen). For variant and SV callers that don't downsample, avoiding reads at super high depth can help provide huge speed improvements on whole genome runs. There are collapsed repeats and single nucleotide runs (all As, all Gs) that can cause a lot of unnecessary compute in regions you won't get useful information from.
I'd like to integrate downsampling as an option but avoid throwing away the high depth reads so the final BAM file is still a full representation of the original fastqs. What would you thinking about making it possible to flag these reads with, say the QC fail check so they're still present in the BAM but will be ignored by variant callers? I'm trying to think up options that don't require having two BAM files (an original and processed) passed around. Thanks for considering this, or for any other options you might think work.
— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/walaj/VariantBam/issues/9, or mute the thread https://github.com/notifications/unsubscribe-auth/AGmfiPALQZ9EPb9xBXed_EH7GAbQVtF0ks5sBng3gaJpZM4Nyfme .
chapmanb
Jeremiah; Thank you for VariantBam, it's a really useful tool and we're looking to incorporate it's maximum coverage filtering into bcbio (https://github.com/chapmanb/bcbio-nextgen). For variant and SV callers that don't downsample, avoiding reads at super high depth can help provide huge speed improvements on whole genome runs. There are collapsed repeats and single nucleotide runs (all As, all Gs) that can cause a lot of unnecessary compute in regions you won't get useful information from.
I'd like to integrate downsampling as an option but avoid throwing away the high depth reads so the final BAM file is still a full representation of the original fastqs. What would you thinking about making it possible to flag these reads with, say the QC fail check so they're still present in the BAM but will be ignored by variant callers? I'm trying to think up options that don't require having two BAM files (an original and processed) passed around. Thanks for considering this, or for any other options you might think work.