HaloPlex for very rare SNV detection (brian: ecc.sh will help?)

[dietmar13]

hello all,

has someone used HaloPlex data (agilent, similar SureSelect) for detection of very rare mutations (down to 0.5%)?

this means very deep data (up to few 100k coverage) for very small regions (~10 k bases). HiSeq 2500 150 PE.

my pipeline will look like this:
- remove adaptors and low quality bases (Q10 or higher?)
-(?) error correct (e.g. with ecc.sh (BBmap)) - BRIAN: do you think this is wise, or will i remove also rare SNVs (the data are up to few 100k's deep)
- map to genome (BBmap)
- realign INDELs and recalibrate quality (GATK)
- call SNVs, using following callers:
--- freebayes
--- loFreq
--- MuTect

suggestions? other callers for very rare SNVs?

thank you,

dietmar

[Brian Bushnell]

dietmar,

I generally do not recommend error-correction if you are looking for extremely rare events. It should not have any impact on indels, but if a SNP is present at a 0.5% rate, it WILL be corrected into the dominant allele with the default settings. The default cutoff is 1/140 (0.7%).

You can adjust this with the "ecr" (errorcorrectratio) flag, e.g. "ecr=400" to correct only events of at least 1-in-400 rarity. Also, there's another flag, eccmaxqual, which sets the maximum quality to correct; if you set "eccmaxqual=25" then bases with a quality value above 25 will not be corrected even if they look like errors. And since you have such high coverage, you can probably be fairly aggressive when quality-trimming, maybe to Q15 or more.

So, I'm not sure if ecc will ultimately be helpful, but if you want to try it, be sure to adjust those two flags.

[dietmar13]

thank you brian,

that is was i thought...