I have worked on metabolomics data and I am not sure the principle of analysis. First is to conduct a PCA. Literature review says it is to test the overall stability of the system. I do not understand it. Then PLS-DA could select some metabolites. Permutation for cross-validation is to check if over-fitting exists or not. I really do not understand the last bit. Why R2 less than 0 means no over-fitting? What if R2 is over 0 on the figure after permutation? I also do not understand how this permutation is done to cross-validate. Could anyone explain it to me? Many thanks.
Hi Cheng, in the metabolomics data you will get many variables and several samples that come from different classes. Sometimes the number of variables >> samples. And from this data, you want to find some important variables so that your research can be conclusive.
PCA help you to understand the natural separation of the sample class. You can say that PCA is a blind classification, but somehow you can put the class legend in it, to distinguish which point belongs to what class. Basically, PCA reduces your variable dimension by creating PCs. PCs are your "new variables." Usually, using PCA your metabolomics data can be interpreted in a very shallow conclusion. Therefore, you can use PLS-DA to do a more robust classification, because PLS-DA is not a blind algorithm. PLS-DA algorithm is a supervised analysis, it already understands your sample class. Sometimes, you'll get a nice separation by using this model compared to PCA.
Also, one of the advantages of PLS-DA is that you can find Variable Important in Projection (VIP). They are the most influencing variables from your metabolomics data. From the VIP you can get a deeper conclusion about your data.
I suggest this R package for more information http://mixomics.org/
Good luck
thanks for your explanation, Fawzan and John. As for the second part of my question, I also do not understand the permutation and cross-validation after PLS-DA. How to interpret the permutation curve, like the attached figure? How to know if it is over-fitting or not? What is the meaning of the figure?
- Badges
- Science topic
- Similar topics
- Linguistics
- Composition Studies
- Publication