Hello,
I have processed a sequence of LC-HRMS chromatogram with my usual workflow without problem.
Then I’ve tried several parameters to do the PCA. It was successful with “Log10 transformation” = yes.
When I run again the PCA changing only “Log10 transformation” = no , only the coordinates of the first dimension of the PCA are calculated. On the metadafile, only the column “PCA_XSCOR-p1” appears, there is no column “PCA_XSCOR-p2”.
I have run this workflow many times, and had no problem before. If I run again the same PCAs (“Log10 transformation” = yes. or no) in an older history where they were successful, it is working fine.
So maybe it is a problem with my files? But I cannot see why.
Could you please help me resolve this problem?
Thank you in advance for your support,
Valérie
Hi Valérie,
Is the ‘Number of predictive components’ parameter set to ‘NA’? If it is the case, then the module suggest auomatically a number of components for PCA. Thus, the number of components selected can vary depending on the effective variability in your dataset. To note, log transformations affect the variability of a dataset, since it has an impact on the variance values that are computed in the process. Thus, depending on your dataset, the automatic number of components selected can happen to change when applying or not the log transformation. This is particularly true if you are using ‘center’ as scaling parameter, and potentially also if using ‘pareto’.
If you want to have specifically the 2 first components, you can always set the ‘Number of predictive components’ parameter to 2. Please be cautious in the potential interpretation you may make of the results, for example by looking at the loading plot to ensure the scaling/transformation you use is relevant regarding your objectives in using PCA, and also regarding the amount of variability that corresponds to the components you are investigating.
Best regards,
Mélanie
Dear Mélanie,
Many thanks for your answer, your suggestion worked just fine: I run again the PCA with the number of predictive component set to 2 instead of NA, and indeed, It gave me the calculation and graphic output in 2 dimensions.
Have a nice day, best regards
Valérie