This is a test version of Biostars. For the public version, visit https://www.biostars.org.
Negative Results in McDonald–Kreitman MK test results

Hi everyone,

I’m running McDonald–Kreitman (MK) tests across several thousand genes to estimate alpha (the proportion of adaptive substitutions).

After filtering out genes with zero values for Dn, Ds, Pn, or Ps, I still observe the following pattern:

  • ~80% of genes are not significant (p > 0.05)
  • Among the significant ones, ~60% have positive alpha and ~40% negative alpha
  • Some alpha values are highly negative (e.g. –24)
  • Alignments are codon-based and appear fine upon inspection
  • Polymorphism frequency threshold = 0.1

I expected a stronger signal of positive selection overall (especially in sex-biased genes), but instead I see a predominance of non-significant and negative results.

My questions:

  1. Is this distribution of alpha (many insignificant, some strongly negative) normal for large-scale MK datasets?
  2. Could alignment quality or population grouping errors produce such negative alpha values?
  3. Are there known biases (e.g., low polymorphism, slightly deleterious variants, demography) that could explain this pattern?

Any insights or experiences with large MK test datasets or codon-based alignments would be greatly appreciated.

Thanks!

mktest
• 1,523 views • link
written by 18d10fee • 18d10fee

1 answer

I will be brief:

  1. Is this distribution normal for large-scale MK datasets?

Yes - ~80% non-significant is typical (most genes neutral/weakly purifying)

  1. Could alignment quality or population grouping errors produce negative alphas?

Possible but unlikely.

  1. Known biases explaining this?

Yes

• 0 views • link
written by Kevin Blighe Kevin Blighe

So you said “possible but unlikely”, but I’m not sure we are talking about the same thing. My concern is specifically about using multiple outgroups in MKtests_JY (https://github.com/jayoung/MKtests_JY/). In my setup, pop1 and pop2 each contain multiple sequences. According to the MKtests_JY code (pop1seqs / pop2seqs), this should be supported. I checked the internal functions and they appear to handle multiple sequences correctly by pooling polymorphisms and counting fixed differences relative to the outgroup(s).

However, I’m still unsure whether having several outgroup sequences could distort the divergence counts enough to cause strongly negative alpha In my case, I’m using 7 sequences as pop1 and 4 sequences as outgroups; but even when I reduce the analysis to a single outgroup, the alpha values don’t really improve. So I’m not convinced the number of outgroups is the issue.

• 0 views • link
written by 18d10fee • 18d10fee

Log in to answer this question.

More posts like this