GATK UnifiedGenotyper -dcov parameter

[liu_xt005]

Could anyone tells what this is exactly controlling?
Some one suggested to use 10 times the average coverage.
Is that correct?
How about variant-calling for multiple samples (BAM)?
Shall we keep the same?
I used -dcov 200.

[ymc]

Supposedly it downsamples the coverage for every site to the dcov value. The default value is 250.

Higher call accuracy can be achieved with higher dcov value. But it also increases the RAM usage. The rule of thumb is 10x the effective mean coverage of your run. So if your run is 40x, any value higher than 400 won't bring you much value.

[liu_xt005]

Thanks a lot to ymc. I changed to 1000, and got a small number of additional variants.

[sirmark]

Have I undestand well -dcov usage ?

If I want consider only reads with a specific coverage for an SNP calling process,
is it userful to use -dcov parameter ?

[Bukowski]

Have I undestand well -dcov usage ?

If I want consider only reads with a specific coverage for an SNP calling process,
is it userful to use -dcov parameter ?

No, it is for *downsampling* positions with greater than this coverage - not considering a specific coverage level.

[sirmark]

can u define me what's a "downsample" ?

[Bukowski]

can u define me what's a "downsample" ?

If you have a position with 500 reads, and your -dcov is 250, a random selection of 250 reads will be used for the genotyping.

So you have 'downsampled' the position from 500 reads to 250.

[liu_xt005]

If you have a position with 500 reads, and your -dcov is 250, a random selection of 250 reads will be used for the genotyping.

So you have 'downsampled' the position from 500 reads to 250.

Therefore the default 250 is a bit low for positions with over 1000 depth. This will be more severe for multiple-sample calling, e.g. you are genotyping 30 individuals.