Gatk3 Vs Dindel For Small Indel Calling

Question

[michealsmith]

I read some posts from Heng Li around two years ago that he suggested Dindel is the optimal software for small indel calling. I'm just wondering if this is still true today because GATK seems to implement Dindel algorithm? Thanks

Answer 1

[lh3]

I guess no one really knows. The performance of Dindel/GATK/samtools varies between data sets and varies with people who are using them. Someone told me samtools is better on their data sets and I have seen posters showing GATK is better. I would say these three callers do not differ too much in accuracy, but probably all of them are way from the optimal at present, so when you compare them, they have very a small intersection.

Broad is experimenting a haplotype caller. It should give better accuracy in the long run, but I have not evaluated it myself. For single high-coverage sample, another viable approach is to use de novo assembly, but it is much more resource demanding and not to the production quality yet.

[michealsmith]

thx Heng, but I remember 2 years ago, you very confirmly said Dindel is the best for small indels because many people have tested and proved that. So the situation has changed? I mean after testing by many more people, now we cannot reach an agreement that Dindel is the best?