Calculating Genotype Likelihood from NGS

[sorento7]

Hi,
I have human whole genome sequencing data (at 30x depth) which is already called using CASAVA. I plan to use these genotype data (>3million SNPs per individual) for quantitative trait association study. regarding the fraction of the polymorphisms which have a low coverage (e.g. < 4) should I filter these out, or it would be better to feed these low coverage data to an imputation pipeline?

[TiborNagy]

Check the mapping quality of the reads. If the mapping quality too low, you can filter out.

[sorento7]

Check the mapping quality of the reads. If the mapping quality too low, you can filter out.

Is that the QUAL field in VCF? so you recommend filtering only based on QUAL and not considering GQ/read depth? i.e. what should be done if quality is good, but depth is moderate/low?

[GenoMax]

Useful thread about filtering VCF files over in GATK forum: http://gatkforums.broadinstitute.org...ring-vcf-files

[sorento7]

Useful thread about filtering VCF files over in GATK forum: http://gatkforums.broadinstitute.org...ring-vcf-files

Thanks for your response. I have gone through the suggestions provided on this link. My specific problem is with variants with good QUAL (hard filter >20~30) but low sequencing depth, which translates to lower certainty about the exact genotype. i.e. there is already some evidence to rule-out REF/REF genotype, but not enough reads to specify REF/ALT vs ALT/ALT.

Is there any recommendation/paper regarding NGS genotyping and association studies? Overall, Does it make sense to filter based on both QUAL and Depth/GQ parameters?