As my field of study is not related to “omics”, I would be more than happy should you comment on the appropriateness of mixOmics package to solve my case below:
My data set is 11 GLCM atributes generated from 10 SEM images of food samples using imageJ software, they serve as input variable, dim(X) = [1] 10 11. For the output Y is the antioxidant activity of the food samples (classified as Low and High). I would like to run PLS-DA to examine whether the samples could be classified by the GLCM attributes. By following your well explained mixOmics vignette I could produce pca, plsda, perf plots. However I am not convident to say something about them.
Kindly advise me whether the mixOmics may be used for the such small samples (the best I could afford unfortunaley) in my experiment setting.
Hi Yohanes,
The methods in mixOmics can be used in an exploratory manner, but we would not recommend we use the perf() function (except perhaps with leave-one-out cross-validation and with a high level of caution regarding the results obtained).
I confirm that the methods are suitable to any data measured on a continuous scale - with the exception of the categorical factor when performing a supervised analysis like PLS-DA.
Kim-Anh
Hi Kim-Anh,
Thank for the reply, here is my first sPLS-DA to show:
Can we say that the predictor variables have sufficient influence to separate individuals?
One more question regarding CIM, when we feed splsda object to cim which values are represented? Thanks.
hi @ykristianto,
Your sample plot clearly shows no separation of your sample groups. You may want to select more, or less variables. Either way, I am afraid to say it is not working well.
The CIM includes the expression data matrices of the selected variables for each sample, with specific aggregation and distances, see ?cim.
Kim-Anh
Hi Kim-Anh,
You’re right, I add more variables and can see the separation. I also drop an outlier. The graph below may be best I could make this time, thinking to do with bigger data set for future. Thanks a lot.
