Python-based
Hierarchical
ENvironment
for
Integrated
Xtallography
phenix.douse is a tool to build water into cryo-EM maps.
1) The water is built into density map also ensuring each water is linked to the protein via H-bond, directly or indirectly (via another water).
2) Atomic model of the macro-molecule is assumed to be as complete and accurate as possible (for example, B factors are refined individually).
3) Water is looked for in a band of 6A width extending out from the protein surface. This allows to build about two solvation layers around protein. Any unmodeled density in this region is treated as potential water.
4) Minimal (dist_min) and maximal (dist_max) distances between any water and protein or other water are 2 and 3.2A, correspondingly.
5) Not all protein atoms are used to assess contacts with prospective water molecules but only those that can participate in H-bonds. These atoms are additionally filtered to leave a subset of atoms that are confidently placed in the map. The confidence is quantified by CCmask (cc_mask_threshold_interacting_atoms) - the correlation between model-calculated and experimental maps for the atom in question. The default is set to 0.5. The CCmask is calculated in a sphere of 2A radius (atom_radius) around the atom in question.
6) Only strong enough density peaks are considered as potential water-candidates. "Strong enough" is defined as the average of map density values interpolated at atomic centers of protein atoms times the attenuation scale factor map_threshold_scale which is set to 0.25 by default.
7) The remaining peaks are filtered by two additional criteria.
One is CCmask (cc_mask_filter) with the cutoff value of 0.5 (cc_mask_filter_threshold). The CCmask is calculated similarly to "5)". Since CCmask strongly depends on atomic B factors and B factors of newly placed water-candidates are not known, unrestrained isotropic B factor refinement is performed for newly placed water prior to CCmask calculation.
Another selection criterion is peak sphericity (scc). The peak sphericity is evaluated as minimal correlation between any combination of two sets of map values interpolated along vectors going out from the peak center and having length of 1A. For a perfectly spherical peak the correlation between any "pair of vectors" (any pair of sets of map values along these vectors) must be 1. The water is rejected if scc is less than 0.5.
1) Geometric (water-protein or water-water distance) and chemical (types of interacting atoms) criteria are rather strong and rigid criteria and do not allow much room for variation. The set default values are taken from standard literature.
2) Map-based criteria, such as various correlations, are more heuristics and rule-of-thumb based. For correlations, 1 means perfect similarity, 0 is no similarity and -1 is anti-similarity. So a choice like 0.5 means some similarity, half-way between none and perfect. Choosing a larger value will mean choosing more confident waters.
3) Map sphericity reflects the observation that water density appears to be approximately spherical most of the time. This observation breaks down in a number of special cases that need a special treatment and that can be a subject of future improvement of the tool. For example, at high resolution water with high directional mobility typically yield elliptical peak shapes. Water coordinated by heavier ions may not even resolved as individual peak or even if resolved as individual peak its spehricity can be largely perturbed.
4) Peak height threshold for selecting peak candidates is another rather ad-hoc criterion. It uses peak heights of confidently placed protein atoms as a guide. However, since water molecules are more mobile and often partially occupied, their peaks are expected to be weaker than peaks of better resolved protein atoms. By how much weaker? That is unknown. Thus some arbitrary attenuation scale is used.
5) Building 'single-atom' entities, such as water or ions, is a much more difficult task to do compared to building more feature-rich molecules such as proteins or multi-atom ligands. This is because it is much easier to confuse a true peak with a noise peak, while a constellation of many peaks (corresponding to protein chain, for example) contains very important information about shape and connectivity, making interpretation such peaks in terms of atomic model easier. Therefore, water building procedure is unlikely to be ever fully automated.
Command line, GUI.
Atomic model (PDB or mmCIF format) and map (MRC format).
resolution is required if cc_mask_filter is set to True.
mode = whole or per_chain: work on the whole model or per chain. Working on the whole model may require a lot of memory.
sphericity_filter = True: use peak sphericity criteria. Disable to add more water.
scc = 0.5: map peak sphericity characteristic. Decrease to add more water.
keep_input_water = False: keep or remove existing water.
dist_min = 2.0: minimal peak-peak or peak-atom distance. Decrease to add more water.
dist_max = 3.2: maximal peak-peak or peak-atom distance. Increase to add more water.
step = 0.3: map re-sampling step.
map_threshold_scale = 0.5: map threshold for peak selection = (mean map value at atom centers) * map_threshold_scale Decrease to add more water.
cc_mask_filter = True: reject water with CC_mask < cc_mask_filter_threshold.
cc_mask_filter_threshold = 0.5: CC_mask threshold for rejecting water.
resolution: map resolution. Required if cc_mask_filter=True
P1 symmetry is the only supported symmetry.
Macromolecular part of the atomic model is as compltete as possible (no missing chains or other unmodelled parts present). The model is fully refined including individual B-factors.
Map resolution is sufficently high to expect water oxygens be visible in the map.
phenix.douse model.pdb map.mrc resolution=2.5
resolution is required if cc_mask_filter is set to True.
https://phenix-online.org/phenixwebsite_static/mainsite/files/presentations/water.pdf
This document reflects the procedure at the time of making the document. The procedure in most current version of Phenix can be (slightly) different.
Pavel Afonine (PAfonine@lbl.gov)