Dear All,
I am performing OPLDS-DA to determine, among my 58 parameters (104 observations), which one(s) drive the separation between my disease group and my Healthy control group. I expect some of the parameters to be highly collinear. And I thought OPLSDA could account for this.
I was working following this tutorial written by Etiene Thevenot: https://www.bioconductor.org/packages/devel/bioc/vignettes/ropls/inst/doc/ropls-vignette.html
when I arrive to:
M.plsda <- opls(dataMatrix, group)
104 samples x 57 variables and 1 response standard scaling of predictors and response(s) 45 (1%) NAs 1 excluded variables (near zero variance) R2X(cum) 0.443, R2Y(cum) 0.218, Q2(cum) 0.0742, RMSEE 0.339, pre2 ort 0, pR2Y 0.1, pQ2 0.05
M.oplsda <- opls(dataMatrix, group, predI = 1, orthoI = NA)
104 samples x 57 variables and 1 response standard scaling of predictors and response(s) 45 (1%) NAs 1 excluded variables (near zero variance) R2X(cum) 0.443, R2Y(cum) 0.217, Q2(cum) 0.0693, RMSEE 0.34, pre1, ort 1, pR2Y 0.05, pQ2 0.05
then I try to set a random test sent by
M.oplsda <- opls(dataMatrix, group, predI = 1, orthoI = NA, subset="odd") Error: No model was built because the predictive component was not significant
Question 1: Is it absolutely not possible to continue computing further the model if Q2Y wis not significant before adding the subset? Question 2: Is there any way I can still try to define the separation model? Question 3: If yes, the variables with higher VIP would or would not have still problems with collinearity?
Sorry if this does not make sense but I am ver new to R and stats.
Thanks so much Adriana.