bcftools mpileup for a set of Illumina+HiFi reads

[sitaramrajaraman]

I have a set of 100 short read bam and 50 long read animal bam files. I want to do joint-genotyping. Would it make sense to run mpileup using pacbio hifi parameters for both? I couldn't find such a scenario in the issues section. We are looking at pop gen analysis downstream.

[tseemann]

@sitaramrajaraman what options of mpileup are used to change the behaviour depending on your read technology?

[jkbonfield]

This is a fair comment. The -X profile options are per machine type rather than mixed data type. I did have plans for interrogating the RG PL field to work out the platform and automatically select the appropriate instrument-specific quality value and data alignment handling options, which would enable it to do this per read rather than per file, but that was never implemented.

I don't know which option would be best I'm afraid.

For @tseemann , try bcftools mpileup -X list. It's only the indel caller which is changing here as this is the bit that really makes a huge difference, although it would be good to get something similar affecting the SNP caller too. Note that by default it uses none of these as the --indels-cns isn't enabled by default. See #2045 for history, discussion and results.

[sitaramrajaraman]

@tseemann , so as @jkbonfield mentioned, the -X parameter in the latest bcftools can handle machine-wise. At the moment, what I have tested is to disable BAQ calculations and use the -q and -Q parameters (-B, -q 20, -Q 20) that could be common between both the illumina and pacbio platforms.

I took a small set of samples for which data was generated using both illumina and pacbio and ran it with these parameters. As a preliminary validation, I ran plink IBD (--genome) to assess sample relatedness and I got ~0.8 between the same set of samples indicating good concordance. I am guessing the mixed platform approach is working but if there are better ways to test cross-platform concordance, I would be happy to try that.

[jkbonfield]

For validation purposes, it's probably best to just pick something like HG002 which will have been sequenced by lots of platforms and produce new mixtures to call. You can then compare it against the published truth sets to generate similar FP/FN plots (filtering by QUAL or other relevant scores).