I'm interested in comparing the intra-regional genetic relatedness between two populations of mice. One population is native to Europe, while the other is native to North America.

I ran PCA to find which population has less genetic structure (i.e. less within population variations). In other words, I'm interested to find which of the two is the younger population (more recent ancestor means more related to each other). The result of PC1 vs. PC2 can be seen below.

PC1 vs. PC2

It is clear that PC1 describes the continental scale, which separates the two mouse populations. The eigenvalue of PC1 is 72.5. Here, the left population is the European mice and the right is the N. American mice. The variations along PC1 for both seemed to be quite similar, except for a few outliers among the right population.

What I'm interested is the PC2 variation. From the plot, European mice exhibited much greater variation compared to the cluster formed by N. American mice. From my knowledge, PC2 is orthogonal to PC1, and it is always the axis with the greatest variation. From this, I concluded that the within population variations among European mice (left) is greater (i.e. more genetic structure) than among N. American mice (right).

I investigated higher components as well. For PC3, the opposite is true. The right population (N. America) exhibited the greatest variation relative to the left population. However, the eigenvalue of PC2 is 3.5, while for PC3 it is 2.2. Which means that PC2 still wins in explaining which of the two populations has the greatest genetic variations. PC4 and beyond doesn't add any new information.

Does my approach seem reasonable to you?

  • $\begingroup$ What do the colors represent? $\endgroup$ – kmm Jul 28 '18 at 23:26
  • $\begingroup$ @kmm: Please ignore the colors. My data are from different sources, it was used for me to check for outliers. The different sources are also from different regions to be able to fully represent a continent. Basically, the left population is European native and the right population is N. American native $\endgroup$ – Rudy Winono Jul 28 '18 at 23:40
  • $\begingroup$ Could you describe the preprocessing of the data? How was data quality standardised across your multiple sources? I don't think ignoring your colours is wise. It looks like a strong batch effect where source is the batch. Do you have data on within and between source variation? Never ignore anything in PCA it is a very powerful method for checking the data had been prepared properly as well as uncovering new information. $\endgroup$ – ReneBt Jul 29 '18 at 6:05
  • $\begingroup$ @ReneBt: I think I didn't explain properly. The colors do represent different regions within Europe. That's why there are different clusters. I believe the pink color is from Britain. The mice are more isolated, hence it differed from Continental Europe ones. Also, I'm not the person who collected the data, but the sources can be trusted. My task is just to analyse the given data. $\endgroup$ – Rudy Winono Jul 29 '18 at 6:10
  • 1
    $\begingroup$ Upwards of 80%of analytics is data quality. If you have verified the integrity of the data by all means proceed. However it sounds like you are passing the buck. This may be an incorrect assumption, but without a lot more detail how you've verified data quality interpretation is impossible to advise on. I've learned to always push back a bit to data providers, I've never met a situation with perfect data, but you need to understand its imperfections to know how to handle it correctly. $\endgroup$ – ReneBt Jul 30 '18 at 7:43

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.