Talk:McDonald–Kreitman test

This article doesn't describe the test
According to McDonald and Kreitman themselves, the McDonald-Kreitman test is a statistical test - to quote the linked abstract of their original article: "Here we propose a simple statistical test of the ...". However, this McDonald-Kreitman test page doesn't actually mention any statistical tests and as such doesn't cover the topic it purports to. What is discussed in the article is a table, a comparison between two proportions and an estimate of α. These comprise calculations relating to a statistical test and are important things people will want to know, but don't constitute a statistical test.

I think an article purporting to describe the M-K test needs to actually describe the eponymous statistical test.

I don't have access to the McDonald and Kreitman article, but according to this article it is a 2x2 Chi-square test performed on the table given in the McDonald-Kreitman test page. By contrast, according to this pdf article McDonald and Kreitman actually use a G-test, which is asymptotically equivalent to the Chi-square test (in most cases they will give the same result and when the numbers are sufficiently large they become identical). The fact that the list of references in the McDonald and Kreitman article include Sokal & Rohlf (Sokal, R. R. & Rohlf, F. J. Biometry, Freeman, San Francisco, 1981) strongly suggests it's actually the G-test, but the difference matters little compared to the complete omission of any test whatever.

My suggestion is to include at the least the following sentence as a new paragraph before the final sentence (the one beginning "As an extension of this ..."):


 * The test itself consists of the G-test performed on the numbers in the table above.

Actually, rather than leave the page in its present state, I'm going to go ahead and put it in shortly, but I'd like someone to confirm by checking the actual article.

Glenbarnett (talk) 08:02, 25 August 2011 (UTC)


 * Changed the article as indicated (but with a little more detail) Glenbarnett (talk) 08:32, 25 August 2011 (UTC)

The reference to the G-test was, for some reason, removed on 8 November 2013. I believe we're back to square one in that this page does not describe the test, namely what statistic should be calculated and what are the critical values to compare it to. Mebden (talk) 17:22, 21 April 2020 (UTC)

Evaluation/Future Edits of the Article
The article does not go into very much depth as to what we can conclude using the Dn/Ds ratio test. I would first dicuss what the null hpothesis is when performing a McDonald-Kreitman test. The null hypothesis all nonsynonymous mutations are expected to be neutral; the ratio of nonsynonymous to synonymous variation within species (Pn/Ps) is expected to equal the ratio of nonsynonymous to synonymous variation between species (Dn/Ds). When comparing the the Dn/Ds ratio to the Pn/Ps ratio, we can use the quantitative numbers of each to determine whether it seems that positive or negative selection is at work here. Look into reading this article for more information: http://nar.oxfordjournals.org/content/36/suppl_2/W157.full.pdf+html?sid=6001f1bd-4744-4bce-8837-293e14e69a3c

Additionally, the article appears to lack any information on the neutrality index. The quantified neutrality index determines the direction and the level of deviation selection has occurred. Try reading an article such as this one to learn more about the neutrality index and its function: http://mbe.oxfordjournals.org/content/28/1/63

I would also add in some information about how the McDonald-Kreitman test underestimates the level of adaptive evolution when slightly deleterious alleles are present. In order to be more accurate, often times when performing this test polymorphisms below a certain number are removed from the data, and the estimate of the level of adaptive evolution is closer to the actual number. This article seemed to be pretty helpful: http://mbe.oxfordjournals.org/content/25/6/1007.short

Caseyafoster (talk) 10:38, 9 October 2013 (UTC)