jeremymcrae
commented
5 years ago Hi Yu,
The principles used in this repository should extend to other species, but the specific implementation probably won't - it's tuned to the data used in the DDD study.
The basic principles are:
- find genotypes present in a child, and not called in the parents
- make sure they are rare in control populations
- make sure the site had sufficient depth in the child and parents to give reliable genotypes
- check they don't occur in segmental duplications
- check for read-level evidence that the site or surrounding gene might be prone to de novo-like errors. This is the tricky part, and relies on having read depths for the reference and alternate alleles, by forward and reverse reads. That allowed filtering out variants with strand biases, or with too many alts in either parent.
For the last part, you could alternatively construct filters that use ref and alt depths, or incorporate genotype quality for something reasonable.
We tried for good sensitivity and specificity, but preferred to be slightly too sensitive rather than specific because a) that suited clinical diagnostics, b) genes showing genome-wide enrichment could have potential de novos validated (and errors excluded), and c) we could calibrate de novo counts by thresholding on DNM quality until the number of synonymous de novos matched expectation.
Hope that helps, Jeremy
ggoodstudydaydayup
Dear, I 'm new for de novo mutation. And I read the paper, Prevalence and architecture of de novo mutations in developmental disorders. And I think it's amazing. Could you please help me to learn more about the DNM filter of this paper? I'm sorry about my lack of knowledge. Thanks in advance. Best wishes, Yu Liu