I modeled a receptor, partitioned the binding sites and docked (ligandfit) a library of small molecules to it.

The results consisted of each of the molecules docked to each partition and a whole lot of the correseponding molecular properties.

I guess the logical thing to do is to do a PCA and reduce the dimensions of the data; I did that and obtained the plot and the coordinates of the molecules.

So here is(are) my question(s), how do i select which the best perfoming molecules according to the new, most important, reduced set of dimensions( the PCs)?

Do i have to do Muiltiple linear regression or Partial least square regression, or are they even necessary, and if they are, how do i do those in DS?

One more thing, Consesus Scoring or PCA in such a scenario as above, which is better, or could they complement each other?

All the help is appreciated

Thanks,

Martis