Inferring heterozygosity from ancient and low coverage genomes
While genetic diversity can be quantified accurately from high coverage sequencing, it is often desirable to obtain such estimates from low coverage data, either to save costs or because of low DNA quality as observed for ancient samples. Here we introduce a method to accurately infer heterozygosity probabilistically from very low coverage sequences of a single individual. The method relaxes the infinite sites assumption of previous methods, does not require a reference sequence and takes into account both variable sequencing errors and potential post-mortem damage. It is thus also applicable to non-model organisms and ancient genomes. Since error rates as reported by sequencing machines are generally distorted and require recalibration, we also introduce a method to infer accurately recalibration parameter in the presence of post-mortem damage. This method does also not require knowledge about the underlying genome sequence, but instead works from haploid data (e.g. from the X-chromosome from mammalian males) and integrates over the unknown genotypes. Using extensive simulations we show that a few Mb of haploid data is sufficient for accurate recalibration even at average coverages as low as 1-3x. At similar coverages, out method also produces very accurate estimates of heterozygosity down to 10-4 within windows of about 1Mb. We further illustrate the usefulness of our approach by inferring genome-wide patterns of diversity for several ancient human samples and found that 3,000-5,000 samples showed diversity patterns comparable to modern humans. In contrast, two European hunter-gatherer samples exhibited not only considerably lower levels of diversity than modern samples, but also highly distinct distributions of diversity along their genomes. Interestingly, these distributions were also very differently between the two samples, supporting earlier conclusions of a highly diverse and structured population in Europe prior to the arrival of farming.