Solubility

[Hiro Sheridan]

I thought this may be useful:
http://pubs.acs.org/cgi-bin/sample.cgi/mpohbp/asap/html/mp7000878.html

[Rajarshi]

On Mar 22, 8:18 pm, Hiro Sheridan <gameshon...@hotmail.com> wrote:
> I thought this may be
> useful:http://pubs.acs.org/cgi-bin/sample.cgi/mpohbp/asap/html/mp7000878.html

Yes, it's an interesting paper that could be a useful way to approach
aqueous solubility predictions. However for the current problem of
methanol solubility, we face a few problems that probably preclude the
use of this approach

* Part of their formulation uses an approximation of \delta G_hydr in
which they employ the log P of the solute. So if we were to use it for
methanol solubility we'd need a value or measure of the (octanol/
methanol) partition coefficient.

* The key aspect of their approach is the use of observed crystal
structures based on which they evaluate U_lattice. I don't think we're
going to easily get crystal structures

However, it should be possible to evaluate terms like \delta G_solv.
It's also interesting to note that their final regression equations
did not involve \delta G_solv or \delta G_hydr - but that might be a
feature of their dataset.

Now, for the problem of methanol solubility, one could evaluate \delta
G_sol by evaluating the difference between the gas phase and solution
phase conformations and see if that correlates to observed
solubilities. However, the paper states that solubilitie dervied from
ab initio calculations of Gibbs free energies showed no correlation
with the experimental values.

As an aside, I did run some randomly selected structures from the 003
set of SMILES through the ensemble models I originally built. And I
have to say that the results aren't too great. One model predicts all
ten new compounds as soluble, another as all insoluble. Given that the
models are based on topological descriptors this is not surprising at
all.

I think the next step is to do some form of free energy calculations
for the solvation step and include that as a descriptor and see if it
helps. Ideally, we'd want to to take into account solid state
features, but I don't think that's going to happen. The downside to
doing ab initio or semi-empirical calculations is that it's going to
be a time sink. If we could get some protocol setup (say using
GAMESS), I could take a look at it on our cluster - but whether it's
practical or not really depends on the timings (the molecules are
relatively big and floppy, so conf searching could be time consuming)