try reparametrisation to optomize scaffold diversity

[UnixJunkie]

use Paretto ranking, BM clustering and play with a.
A change of the kernel function might also bee needed but this would have a huge impact
on performance.

[UnixJunkie]

related to #6

[UnixJunkie]

We should play with a (the dx parameter in fact), weights of each space (steric, hydrophobic and
electrostatic). Let's not change the kernel function or it will become too slow.
Issue #6 is more important since it can
improve perfs in several chemical spaces.

[UnixJunkie]

the effect of the cut rot. bounds option should be studied too

[UnixJunkie]

go in the p4 space for this. Also, go out of the molecule like one H2O shell away.