Hi,
I have been using PAM to derive a methylation gene signature between classA and classB. I have 262 samples for classA and 155 samples for classB, and a total of 3035 probe values. For training set of PAM I randomly shuffled the two datasets(A and B), and have taken ~50% of total samples in each class into the training set. So the training set has 131 and 78 samples for classA and classB respectively for 3035 probe values. The problem is if I repeat PAM for the same training set, i get different number of significant probes and misclassified samples both times. I am not understanding how that is possible if the training set remains same. Also, if I get same number of significant probes and misclassified samples at different threshold values after cross validation, which threshold should I be using? the one that is less? or the one that is more. The error rate remains same at different thresholds in this case, but then which one should be considered for further analysis?
Your help is appreciated. Thanks!
