Hi, I'm looking at the calculations of Malinsky et al (2020), where they detail the logic behind Dsuite's calculations. I have a mixed haploid and diploid species dataset where there is clear signs of hybridization. This appears to be supported when running the Dtrios command (highly significant D values and elevated f4-ratios). In my vcf dataset I have mixed haploid and diploid specimens. I suspect this presents a problem for calculating values, particularly if allele frequencies are biased due to the haploid nature of the data (i.e. rather than a potential 0.5 allele frequency, a single haploid individual forces that value to 1 or 0); Dsuite has not offered any issues. In particular, the calculation of the f4-ratio splits derive allele frequencies in P3 into two subsets for each SNP site, which can be done with biallelic data. How then am I getting values with a single haploid P3 specimen? Am I overthinking this or does this have a significant impact on calculating those values, and hence my interpretation of hybridization and admixture proportions?
Thanks in advance!
Hello, I've not seen a SNP-caller that will call genotypes by specifying some samples as haploid, others diploid. Are haploid genotypes coded as 0 or 1 in your VCF, while diploid genomes as 0/0, 0/1, or 1/1? Or are haploid coded as diploid (0/0, 1/1), but never 0/1?
It seems to me that If there is no inbreeding/selfing, then allele frequencies shouldn't depend on whether you have diploid or haploid samples. If there is inbreeding, then I could see where there could be biased estimates of allele frequencies.
Ultimately, to understand how Dsuite is calculating the D stats in your case probably requires either sorting through the source code or performing a toy example where you manually calculate the expected results.
Hi Jonathan, thank you for the response. I used bcftools, for which a samples.txt file can be specified, indicating what .bam files to compile and the ploidy level of each sample. So genotypes are coded as 0 or 1 for haploids, and */* for the diploids.
I hadn't considered the potential complications of inbreeding, that is certainly a potential concern here. I'll see where I get with your suggestions. Thanks!
Yes, you can but to avoid bias it is better not to, as there may be unknown, unseen consanguinity in the population
Thank you again for the responses. I'm hoping someone can clarify a couple points, or challenge my reasoning here. Where I am analysing hybridization between species, I expect many of the variant sites in the haploid specimens represent homozygous alleles in diploids (fixed differences among species); for the most part, I expect allele frequencies are unaffected by ploidy level, simply because I am working at the level of species. While a caveat of the analysis, I therefore do not expect a couple haploid specimens to greatly alter the overall signal. Were this analysis at the population level, I would be more concerned, and take steps to remove haploids from the analysis.
Regarding inbreeding, this is expected to diminish heterozygous alleles in individuals, but not the overall allele frequencies in a population. I therefore expect the bias introduced from inbreeding can be treated by incorporating multiple specimens into the calculation of allele frequencies (something Dsuite is able to do).
It seems to me that the problem with inbreeding in your attempt to calculate allele frequencies might be that inbred samples have a higher probability of contributing 2 of the same allele (i.e., they have a higher probability of being homozygous then expectations under random mating). That is not generally a problem for estimating allele frequencies. However, it strikes me that if you have a mix of diploid and haploid individuals, and different populations/species have more or fewer diploid samples, then those populations with more diploid individuals might be biased toward inflating the frequency of the common allele. I could be wrong, but that's my rational.
Another way to do the D-test would be to use a single individual from each population/species and then draw only a single allele at each site from each taxon. This is in fact how the test was first conducted by Greene et al. 2010 Neanderthal paper.
You might run the test using only diploids, only haploids, or only one of each and make sure your result is robust to the sample choice.
- Badges
- Science method
- Similar topics
- Biological Science
- Ecology