Hi Andrew,
I've been helping a colleague with a dataset at ~3.8 Angstroms (phases and map from AutoSharp using SAD phasing). At the moment he has an incomplete polyALA model consisting of two chains related by NCS and a few extra sections. Refinement seems to proceed OK using phenix.refine version 1.4-3, but the resultant maps have blobs of negative density in the core of the protein. I am assuming that these are a side-effect of the bulk-solvent correction part of the refinement which is confusing the region where the protein sidechains would be as solvent. There are no corresponding negative peaks on the surface of protein, but some positive peaks where larger sidechains may be.
It could be the footprint of bulk solvent mask but well could be something else. At this rather low resolution and very incomplete model you are likely to expect some map artifacts. Anyway, it costs nothing to try / check the following: - The was a very similar issue brought to phenixbb by Morten Grøftehauge; - You can try running phenix.refine with "optimize_mask=true" option that will optimize solvent and shrink truncation radii; - You can place dummy atoms with occupancy=0 into negative density blobs so the mask is not set there - If I remember correctly, this is what worked out for Morten. - Which maps are you looking at: "filled" or not "filled"?
Is this interpretation likely to be correct and if so, which parameter should be altered in the Bulk Solvent section of the input to allow for a bigger mask around the protein. I see that it isn't recommended to change the parameters, but the refinement.mask.solvent_radius=1.0 looks like an obvious candidate.
This is one of the reasons why phenix.refine has so many parameters exposed to users so people can change them if they want. The only thing that is not recommended is changing the parameters without clear idea why do so and what the parameters being changed mean. Pavel.