I’m working with a tetrameric ion channel. After NUR (res 4.10) I did C4 sym expansion followed by local refinement (res 3.7) to model a single protomer of the channel, as the NUR volume quality was lacking in the TMD and the peripheral regions of the protein. I built the best possible model I could with the map and according to Phenix (CCmask: 0.72) and the PDB validation tool statistics.
To build the NUR C4 map, I took the protomer, copied it four times, and docked it into the NUR C4 map. After very minor tweaking in ISOLDE to get the chains into the exact right positions I ran RSR in phenix and overall have a good model (99% rama favored, no rotamer outliers, good CaBLAM etc) but the CCmask has gone down to 0.66, and RCSB validation shows Q scores per chain of around ~0.285.
This makes sense to me. Portions of the NUR map that are not resolved well (like blobs or amorphous shapes) or are lacking in density are going to impact CC and Q scores. The local refine map has great contouring of sheets and helices, whereas the NUR map doesn’t always have these in the poorly resolved regions (hence the need for a focused local refinement).
My initial thought was I needed to rectify this by deleting sidechains that are not in density, or even remove the small segments (sometimes up to 5+ amino acids) of chains that don’t seem to be in an appropriate density. I don’t have a lot of problematic areas, so it’s not a difficult exercise. However, after looking at some other models of similar channels in the PDB, I see this isn’t necessarily done. I’ve found several examples where there are focused refinements of the deficient areas, but the full map+model have models that are built into no density or problematic density.
I don’t know if there is a consensus on what to do in these situations, and I can see the very valid pros and cons of each option investigators choose to do. Composite maps also seem available, but with lots of strong opinions in the field.
Does one just build a full model using the well reconstructed and validated protomer, and ignore the deficiencies of NUR map? Or should you try and make the protomer model match the map and fix deficient segments through pruning and removal?
Working with symmetry expanded local refinements in C1 for C4 transmembrane regions the local refinement maps are very subtly shifted, such the LR needs to be fit back into the C4 NUR prior to multiplication, this is true for non-symmetry maps also where fit the LR back into NUR. Such is the case for making a C4 composite map: 1) multiply LR by mask in chimerax, 2) fit masked LR into NUR, 3) maximize masked fit LR x3 and rotate around the Z axis (90/180/270) and 4) maximize the 4 together. Without fitting the LR in to the NU there often issues such as side chains slightly shifted that are noticeable when making a composite or when apply symmetry to the model that was directly built in LR (this seems to be your case). Fitting the LR into the NUR shifts the map a minor amount that removes these issue such that the C4 will match up for any side chains you can see. If you fit a map multiple times it may be slightly different shift, so check how well it fits in all the rotated densities.
I prefer composite maps, as seen above they require a detailed investigation of the interaction points between the LR maps and comparison to the C4 and C1 original maps to validate. If your case the a NUR 4.1A vs 3.7A LR sounds significant and composite would likely present the data better and the LR sounds necessary for sidechain information for impactful regions. The large difference between C4 NUR and C1 sym exp LR needs looking into for what parts improve and how tetrameric was the map originally.
As to keeping residues/side chains, its what does density support is the ideal. The murky bit occurs when you don’t have good resolution of important regions, such as the pore may be nicely resolved yet what amount of filtering/postprocessing is required to see density for those peripheral helices. Yet for a protein with X known TMs it would look odd if there say there are X-4 TMs depicted in model, are those 4 peripheral TMs there or just mobile? Likely to get more questions about why your missing those helices then putting them in, such adding in those 4 TMs to the model and letting the validation metrics assuage it seems to be the case. From a structural standpoint doesn’t seem to be the correct method to go about it, yet to present the data to non-structural groups which is a large proportion of the individuals this seems like the better method.