What is the distribution of indel sizes in a healthy human genome? of insertion:deletion ratios?

My understanding is that indels are from 1bp to 10Kb, and a healthy genome has ~400K-500K Indels. Surely most of these are small.

What is the distribution of insertion sizes in a healthy human genome? What is the distribution of deletion sizes? What is an average ratio of insertions:deletions?

One of the 2015 papers from the 1000 genomes project has a nice figure (figure 1) showing the size distribution of medium to large sized insertions and deletions:

From another 2015 1000 genomes paper, one can see that the absolute number of smaller indels is much larger, though an exact size range isn't given (as far as I saw). If you really want to know that, just download the most recent 1000 genomes VCF file(s) and compute the exact median size and/or distribution that you want.

Genome-In-A-Bottle (GIAB; version 3.3.2) contains 3.21M SNPs on auto+X chromosomes and 0.51M INDELs in 2.58Gb confident regions. The ins:del ratio is 0.92. On the CHM1-CHM13 pacbio assembly (European ancestry), there are 3.57M auto+X SNPs and 0.58M INDELs in 2.71Gb confident regions [reference] with an ins:del ratio 0.99. These give you an idea of relative short INDELs.

For indels longer than 50bp, I would highly recommend to read this paper by Chaisson et al. This call set is constructed from the whole-genome assembly of the CHM1 genome. It is far more comprehensive and probably more accurate than all the other call sets. The following is Table 1 from the paper:

I have to say that the high ins-to-del ratio is surprising to me, but this is what the data tells us. I have done a similar analysis and reached a similar ratio.

PS: These numbers are essentially from one haplotype. They will be higher for a diploid sample.

EDIT: some numbers in the original post are wrong.

• "GIAB ignores hard regions that tend to harbor more indels" -- that's interesting. Do you have a source for that or could you explain more? – Randoms Jul 19 '18 at 3:28
• @Randoms Sorry. Some of my numbers are wrong. Please see the update. As to your question, GIAB excludes long STRs, while half of indels come from STRs. You can see that "(0.58-0.51)/(2.71-2.58)=0.54" indels per kb outside the GIAB confident region is much higher than "0.51/2.58=0.20" inside its confident region. – user172818 Jul 20 '18 at 1:35
• Ok thanks. Where are those numbers in that coming from? – Randoms Jul 20 '18 at 4:04
• @Randoms These VCFs are public. You can count by yourself. – user172818 Jul 20 '18 at 15:56
• Could you explain where the numbers for the CHM1-CHM13 dataset are coming from? The Syndip paper you linked refers to "0.38 million 2–50bp INDELs". The rep2.37.broad.hc.raw.vcf.gz VCF, by my count, has 936587 indels—493900 are within [2, 50]. The full.37m.vcf.gz VCF has 1145454 indels—556256 in [2, 50]. – Randoms Jul 23 '18 at 4:32