Can I use chi-squared test of independence to compare base compositions of human genome?

Question

I need to prove that a given region on the genome maintains has the same base composition proportions (20% A, 15% T, 19% C, etc…) of that genome. I thought about doing Chi-squared test of independence since I deal with exact counts. As a result, I found $p<0.05$ suggests that they are independent so I say the base composition was changed. I just wanted to get a review of my action.

mydf:
                hg19      AR
            A 854963149  705448
            T 856055361  705519
            G 593325228  482498
            C 592966724  479646
            N 239850802 2806685

chisq.test(mydf)

Pearson's Chi-squared test

data: mydf X-squared = 15718000, df = 4, p-value < 2.2e-16

Is this method valid ?

Answer 1

Technically what you have done is fine; a chi-square test is appropriate for this type of count data in defined categories.

The problem is interpreting the result. N represents an unknown base in the genome data alignment, not a specific base (A,T,C, or G). Over the whole hg19 human genome of 3.1 billion bases there are only 7.6% N (unknown) base calls, but your AR region has over 50% N, the major contribution to your highly significant chi-square statistic. AR must be a very hard to sequence region.

I certainly wouldn't include N calls in an estimate of "base composition," which biologically should only consider ATCG calls. It's impossible to know what the true A, T, C, G values are for over half of the AR region; the unknown positions might, say, be biased toward high C/G composition. So I don't know that you can come to any reliable conclusion about the "base composition" of region AR except that it's much harder to sequence reliably than most of the human genome.