Re: [phenixbb] Fo-Fc omit map for ligand in the protein-ligand structure

Absolutely! But the bias is against the hypothesis that you so desperately want to be true. If that map still showed density that convinced you that something is binding, that is strong evidence. Remember, this is the discovery map so it must have been convincing despite its flaws. If you could only "discover" the ligand by assuming its presence and then seeing its density you have a problem. If you built a ligand there must have been a map that convinced you. It may have been the original MR map where that ATP was so clear that you built it from the start. It may have been a map calculated from a nearly complete refinement of everything else whose precision allowed the visualization of some weak binding ligand. Whatever it was, there had to be such a map in the history of the project. If that map convinced you it should be convincing to the reader (and reviewer) otherwise you are not being skeptical enough. It is hard to calculate phases from a map without making an affirmative statement about the density at every point. The FFT doesn't have an input value for "I don't know". You could calculate a bunch of FFT's with different densities in this region and build a PDF in reciprocal space. Of course the PDF's for each reflection will be tightly coupled to all the others and no one really knows how to represent that. But it isn't the case that we know nothing about the density. We know it is not a uranium cluster and we know the density is not negative. In fact we have pretty specific ideas about what it there, bulk solvent, vacuum (yes, some cavities have nothing in them), or a ligand. Now we are into hypothesis testing - which model works best. If you assume vacuum (as a prior) and density comes back or you assume bulk solvent and weaker density comes back one would doubt the prior in each case. If vacuum is eliminated then you have to have a test between bulk solvent and a ligand (and we don't know which ligand yet). The difference is that the density for a ligand will have high resolution components (say >4 A) where bulk solvent will not. Now we want to say, what is the smoothest map we can construct that fits the data? This sounds like Bricogne's maximum entropy maps. If the data does not demand high resolution features you should not presume they exist and should not build a ligand. Dale Tronrud On 9/16/2014 9:22 AM, Pavel Afonine wrote:

Hi Dale,

your points are all very good!

Let's have a look at this at a different angle though. Bulk-solvent is part of the model: it is the model for the other "half" of the unit cell volume that is not interpreted in terms of individual atoms.

If ligand is not built yet this means the program will use an incorrect model (bulk-solvent) to describe ligand's region. Isn't it introducing a bias?! I would think so.

Pavel

On 9/15/14 11:41 AM, Nathaniel Echols wrote:

...

On Mon, Sep 15, 2014 at 8:58 AM, Dale Tronrud mailto:[email protected]> wrote:

An alternative you might want to consider is what I call the "discovery map". At some point in the refinement process there was a map that convinced YOU that this ligand was present. You should be the hardest person to be convinced so that map will be both an omit map (because the model had been refined without the ligand prior to this) and clear enough to satisfy the reader.

Agreed - this is much easier than placing the ligand and then having to run SA. I suspect this is a case where "oral tradition" has been misinterpreted, since everyone is told that the SA omit map is the "standard" for showing unbiased difference density, and they don't realize that it's better to simply avoid the bias to begin with.

-Nat

_______________________________________________ phenixbb mailing list [email protected] http://phenix-online.org/mailman/listinfo/phenixbb