| 1 |
The view command of bcftools calls variants, tests Hardy-Weinberg |
| 2 |
equilibrium (HWE), tests allele balances and estimates allele frequency. |
| 3 |
|
| 4 |
This command calls a site as a potential variant if P(ref|D,F) is below |
| 5 |
0.9 (controlled by the -p option), where D is data and F is the prior |
| 6 |
allele frequency spectrum (AFS). |
| 7 |
|
| 8 |
The view command performs two types of allele balance tests, both based |
| 9 |
on Fisher's exact test for 2x2 contingency tables with the row variable |
| 10 |
being reference allele or not. In the first table, the column variable |
| 11 |
is strand. Two-tail P-value is taken. We test if variant bases tend to |
| 12 |
come from one strand. In the second table, the column variable is |
| 13 |
whether a base appears in the first or the last 11bp of the read. |
| 14 |
One-tail P-value is taken. We test if variant bases tend to occur |
| 15 |
towards the end of reads, which is usually an indication of |
| 16 |
misalignment. |
| 17 |
|
| 18 |
Site allele frequency is estimated in two ways. In the first way, the |
| 19 |
frequency is esimated as \argmax_f P(D|f) under the assumption of |
| 20 |
HWE. Prior AFS is not used. In the second way, the frequency is |
| 21 |
estimated as the posterior expectation of allele counts \sum_k |
| 22 |
kP(k|D,F), dividied by the total number of haplotypes. HWE is not |
| 23 |
assumed, but the estimate depends on the prior AFS. The two estimates |
| 24 |
largely agree when the signal is strong, but may differ greatly on weak |
| 25 |
sites as in this case, the prior plays an important role. |
| 26 |
|
| 27 |
To test HWE, we calculate the posterior distribution of genotypes |
| 28 |
(ref-hom, het and alt-hom). Chi-square test is performed. It is worth |
| 29 |
noting that the model used here is prior dependent and assumes HWE, |
| 30 |
which is different from both models for allele frequency estimate. The |
| 31 |
new model actually yields a third estimate of site allele frequency. |
| 32 |
|
| 33 |
The estimate allele frequency spectrum is printed to stderr per 64k |
| 34 |
sites. The estimate is in fact only the first round of a EM |
| 35 |
procedure. The second model (not the model for HWE testing) is used to |
| 36 |
estimate the AFS. |