1
$\begingroup$

I find that ~70% of the variance is distributed between the first 5 principal components.

enter image description here

I am guessing that this is not the right analysis to cluster variables into new features. However, there is still a clear clustering if I plot PC1 vs PC2, PC2 vs PC3, PC3 vs PC4 and PC4 vs PC5.

enter image description here

My experience doing PCA is quite recent and I would like to get help on how to interpret these results.

Should I drop further analysis (i.e. feature selection) and switch to another approach, and if so, what would you suggest?

Thanks!

$\endgroup$
  • $\begingroup$ How many variables you used for PCA? From the first figure it seems that they were 7. In that case, having 5 components does not allow for a "big" dimension reduction of the initial variables. If your aim is to classify cases, it seems to me that a cluster analysis is applicable to the data you are studying. Apparently that was already done since the 4 scatter plots already have the colouring on the clusters. What about doing several ANOVA, one for each variable (I'm assuming there are 7 variables), with the categories defined by the clusters. Can you please state your research question? $\endgroup$ – Ertxiem Mar 27 at 14:30
  • $\begingroup$ Hello @Ertxiem. I see that my qustion was not properly formulated. I have 60 individuals. They come from 6 different populations (10 indivuduals/population). For each individual I took >5Mill measurements called SNPs (SNP1 ...SNPn). The scatter plots show one dot for each individual. They cluster according to the population they come from. I want to identify the SNPs that have the largest contribution in forming these clusters. $\endgroup$ – Sergio.pv Mar 28 at 9:47
  • $\begingroup$ In principle, the variable that have the largest correlation with the 1st factor is one candidate. The next candidate could be the variable with the largest correlation with the 2nd factor. Nevertheless, the ANOVA tests can also help you with that. $\endgroup$ – Ertxiem Mar 28 at 15:33

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.