[hail] implement KING #9343
[hail] implement KING #9343
Conversation
danking
force-pushed
the
king2
branch
2 times, most recently
from
4acf83d to
8b026c4
Compare
|
Some shots of the rendered documents. Methods
Relatedness
King
|
|
Needs a rebase to address conflicts. |
|
|
||
| - :math:`i` and :math:`j` be two individuals in the dataset | ||
|
|
||
| - :math:`N^{Aa}_{i}` be the number of heterozygote genotype force individual |
There was a problem hiding this comment.
Is this right? "heterozygote genotype force individual" sounds wrong, but I don't know the math here. At least should probably be "individuals"? I don't know what the "force" part means.
There was a problem hiding this comment.
should be "genotypes for"
| 4 p_s (1 - p_s) (1 - 2\phi_{i,j}) | ||
|
|
||
| This identity reveals that the quotient of the expected genetic distance and | ||
| the four-trial binomial variance in the allele frequency, represents, |
There was a problem hiding this comment.
Don't think there should be a comma between "frequency" and "represents".
CHANGELOG: Implement the KING method for relationship inference as hl.methods.king. Just look at the last commit. The other commits are PRs that I hope will merge on Tuesday. This PR implements `hl.methods.king` a new, relatively fast, method for relationship inference on genotype data. I am eager for criticism of the "Notes" section in which I attempt to describe the KING method to a Hail user with only a basic understanding of genotype matrices and Hail. I also include a benchmark which exercises MT->BM, matrix multiply, and BM->MT. We have an opportunity for a substantial improvement in performance by BM->replacing the BM interface by one which permits multiple entry fields. In BM->particular, note that I have to convert from row-partitioning to BM->block-partitioning four times! Co-authored-by: Gonzalo Anyosa <ganyosagalvez@gmail.com> Co-authored-by: Ahmed Benghomari <abenghomari@gmail.com>
| datasets containing one homogeneous population. :meth:`pc_relate` is appropriate | ||
| for datasets containing multiple homogeneous populations and admixture. | ||
| Hail provides three methods for the inference of relatedness: PLINK-style | ||
| identity by descent, KING, and PC-Relate. |
There was a problem hiding this comment.
Maybe these method names should be links to their papers / websites?
There was a problem hiding this comment.
I added references.
| identity_by_descent | ||
| pc_relate | ||
| king |
There was a problem hiding this comment.
Nit, but let's put them in the same order as above (ibd, king, pc_relate).
| identity_by_descent | ||
| pc_relate | ||
| king | ||
|
|
||
| .. autofunction:: identity_by_descent | ||
| .. autofunction:: pc_relate | ||
| .. autofunction:: king |
There was a problem hiding this comment.
Let's match the order we describe them in (which is also alphabetical)
| - :math:`X_{i,s}` be the genotype score matrix. Homozygous-reference | ||
| genotypes are represented as 0, heterozygous genotypes are represented as | ||
| 1, and homozygous-alternate genotypes are represented as | ||
| 2. :math:`X_{i,s}` is calculated by invoking | ||
| :meth:`CallExpression.n_alt_alleles` on the `call_expr`. |
There was a problem hiding this comment.
Is this a matrix or a vector? (Wondering because it refers to a particular sample, which makes me think of one column of a matrix table). What is little s?
There was a problem hiding this comment.
It's a matrix. s is for SNP.
There was a problem hiding this comment.
I added a sentence describing the subscripts.
| d_{i,j} = 4 N^{AA,aa}_{i,j} | ||
| + N^{Aa}_{i} | ||
| + N^{Aa}_{j} | ||
| - 2 N^{Aa,Aa}_{i,j} |
There was a problem hiding this comment.
In the images you posted, the tops of the superscripts look cut off
There was a problem hiding this comment.
So, this is a problem with any symbol that has both subscripts and superscripts. I found the root-cause and proposed a fix to the sphinx-katex people: hagenw/sphinxcontrib-katex#33.
I'm inclined to wait for them to fix rather than try to monkey patch this. I'm not sure that my fix doesn't cause issues on other browsers (though I did verify our website looks fine on a phone).
| Notes | ||
| ----- | ||
|
|
||
| The following presentation summarizes the methods section of Manichaikul, |
| # We need the count of times the pair is AA,aa and aa,AA. da_ref_var is only AA,aa. | ||
| # Transposing da_ref_var gives da_var_ref, i.e. aa,AA. |
There was a problem hiding this comment.
What does da_ref_var refer to? Old name for ref_var?
This reverts commit 1f1abcb.
CHANGELOG: Implement the KING method for relationship inference as hl.methods.king.
Just look at the last commit. The other commits are PRs that I hope will merge
on Tuesday.
This PR implements
hl.methods.kinga new, relatively fast, method forrelationship inference on genotype data. I am eager for criticism of the "Notes"
section in which I attempt to describe the KING method to a Hail user with only
a basic understanding of genotype matrices and Hail.
I also include a benchmark which exercises MT->BM, matrix multiply, and
BM->MT. We have an opportunity for a substantial improvement in performance by
BM->replacing the BM interface by one which permits multiple entry fields. In
BM->particular, note that I have to convert from row-partitioning to
BM->block-partitioning four times!