In the component matrix, variables with higher value indicate a higher contribution of that component in that particular PC. However, in my case, where my variables are hundreds of lipid compound, I found that there is no higher value observe for all component in the matrix. The highest value observed is 0.29 and there many component with more or less the same value. How should I explain this result? Is this due to the large number of variables that I have? Or is this some sort of indication that my analysis is wrong? Thanks!!
You should have used some of the feature selection (subset selection, variables reduction) methods before PCA.
Hi Anders,
I want to find a pattern/clustering in my samples studied (and yes, I do find clustering) and then find the most important compounds that cause the clustering.
Ili,
It turns out that interpretation with PCA is not that simple. I have been working on that for very long now, and I recommend you use this method:
http://amstat.tandfonline.com/doi/full/10.1080/10618600.2016.1265527
Another thing I commonly do is to select the clusters of observations and apply omeda to identify the variables (compounds in your case) related to them:
http://wdb.ugr.es/~josecamacho/research/contribution.php?key=omeda
Also, you could use coclustering, but not sure in which software package you will find it.
Best,
Pepe
Is it molecular simulation data? Have you looked if the variables are fairly normal distributed before applying PCA?
PCA is fairly robust against departures from normality. As a matter of fact, it is comonly used to detect outliers or trends with really non-normal data.
- Badges
- Science topic
- Similar topics
- Agriculture
- Soil Science