Comparing results from ADMIXTURE and EIGENSTRAT

Question

[misteryxdxdxd]

I am new to this so please bare with me. I have the results from ADMIXTURE which looks something like this

BC001 ABC2CGM1295    0.389581 0.562372 0.048047
BC002 ABC1CCR1227    0.550950 0.445775 0.003275
BC003 ABC2CGM1213    0.407749 0.568932 0.023319
BC004 ABC1CCR1274    0.611861 0.367469 0.020670
BC005 ABC2CGM1254    0.567774 0.431889 0.000337

...rest omitted...

Where the columns are family ID, individual ID, CHB ancestry, JPT ancestry and YRI ancestry.

And I also have the pca.evec from eigenstrat that looks something like this

           #eigvals:    40.900    20.237     1.241     1.484     1.137     1.128     1.124     1.123     1.120     1.119     1.116     1.115     1.113     1.123     1.134     1.106     1.105     1.103     1.103     1.098
   BC001:ABC2CGM1295     0.0157      0.0326      0.0170      0.0194     -0.0007     -0.0006     -0.0095     -0.0121      0.0068      0.0019      0.0001      0.0041      0.0045     -0.0002     -0.0013      0.0006      0.0033     -0.0160     -0.0124     -0.0061             Case

...rest omitted...

I want to look at whether the YRI ancestry calculated from admixture is similar/different from that calculated from eigenstrat. I can clearly see the YRI ancestry in the admixture results, but how can I get that information from the pca.evec results? All I have are the individuals with their respective eigenvectors. How can I find out what their YRI ancestry are based on that?

Answer 1

[eric.kern13]

First, a couple of things that may have gone wrong:

• There is no natural association between the first eigenvector and the first person, the second eigenvector and the second person, etc.
• Why does that comment say eig vals if you're talking about the individuals' eigenvectors?

Now my answer. In a study with n subjects and p genes, EIGENSTRAT produces eigenvectors of length p. You want a vector of length n; one entry per subject. The calculation you should do: take the top eigenvector and perform a dot product with each person's original column of SNPs. That's the number you want for that person. You may also want to look at the second eigenvector.