I was wondering if anyone could help me with a couple of things:

1. I am running k-means for 3000 genes, and realized that everytime that I run with the same data set, different clusters with different amount of genes are formed. I concluded that my expression profiles are not clear cut and therefore genes jump from one to another cluster every time because of the character of the algorithm.. However if I run hierarchical clustering before K-means, somehow I get the same result always???..which does not make much sense to me, since both analyses are sepaparated and should not influence each other.
2. Therefore I am trying the k-means support as a way to give some statistical support to the clusters formed. However I am confused on what is the exact difference between the number of k-means runs and the number of interations in the parameters...