# 7. 3D Map preprocessing

The Fig. 7.1 shows the workflow that we are going to detail in this section.

Fig. 7.1 Scipion framework detailing the workflow generated after 3D map preprocessing.

## 7.1. Map sharpening

As we have indicated before, since map sharpening contributes to increase signal at medium/high resolution, we recommend to perform this map preprocessing step before tracing the atomic model of cryo-EM 3D maps . To accomplish this task, a couple of automatic alternatives are available in Scipion: 1) local sharpening method independent of initial model, based on local resolution estimation ( (Appendix Local Deblur Sharpening)), 2) deep learning-based sharpening approach ( (Appendix DeepEMhancer Sharpening)). Although both sharpening methods display good results, these are not identical but complementary since LocalDeblur maximizes specially details like the secondary structure, whereas DeepEMhancer maximizes connectivity, favoring the fair tracing of the molecule skeleton.
Although the common first rule in both sharpening strategies is counting on half maps to get the best performance of the methods, or the average raw map otherwise, exceptionally in this case, to illustrate the procedure we are going to use the final postprocessed map deposited in the database, where no half maps have been submitted together with the final map.

### 7.1.1. Sharpening with LocalDeblur

Since LocalDeblur takes advantage of map local resolution to increase the signal, we have to compute this local resolution as first step to apply the LocalDeblur sharpening method. Although different algorithms could be used to compute local resolution, we have selected MonoRes , implemented in in the protocol xmipp-local MonoRes (Appendix Local MonoRes).
Since a map binary mask has optionally to be included as a parameter in this protocol, we will build a mask by using the Scipion protocol xmipp-create 3d mask (Appendix Create 3D mask) as starting step in the local resolution estimation process. Open the protocol form (Fig. 10.7 (1)) and fill in the tap Mask generation (2) with the input volume (3) and the density threshold (4). By default, the level value observed in ChimeraX main graphics window (Fig. 6.3) Tools → Volume Data → Volume Viewer → Level can be selected as threshold. In the Postprocessing tap (Fig. 10.7 (5)), select Yes in Apply morphological operation (6) and maintain the rest of options by default. After executing this protocol (Fig. 10.7 (7)), the morphology of the mask generated can be checked in slices by clicking Analyze Results (8).

Fig. 7.2 Filling in the protocol to create a mask of the initial volume.

ShowJ, the default viewer, allows visualize the mask with shape similar to the starting volume (Fig. 7.3).

Fig. 7.3 Visualizing the mask of the initial volume.

NOTE: In case you would like to use a previous computed mask, you can do it simply by importing it using the protocol import mask (Appendix Import mask).
Once the mask of the starting map has been created, the protocol of xmipp-local MonoRes can be completed to get the estimation of local resolution. Open the protocol (Fig. 7.4 (1)) and include the starting map (2), as well as the binary mask (3). Finally, based on the map resolution (3.2 Å), select the default resolution range between 0.0 and 6.0 Å (4).

Fig. 7.4 Completing the protocol to estimate the local resolution of the metHgb map.

Execute this protocol (Fig. 7.4 (5)) and analyze the results (6). The menu of results (Fig. 7.5 (A)), among other views, shows the histogram of local resolutions (1) and the resolution map in (2). The histogram of resolutions, which displays the number of map voxels showing a certain resolution, allows to conclude that the majority of voxels evidence a resolution between 3.2 and 3.5 Å, quite close to the published map resolution (3.2 Å). The resolution map shown by ChimeraX details the resolution of each voxel (Fig. 7.6). The bar on the left indicates the color code for resolution values.

Fig. 7.5 xmipp3-local MonoRes menu of results (A) and histogram of resolutions (B).

Fig. 7.6 Resolution map in ChimeraX.

Local resolution values of the input map allow to compute the sharpened map by the xmipp3-localdeblur sharpening protocol, which implements an iterative step descending method that doesn’t require initial model. To run this method, open the protocol (Fig. 7.7 (1)) and include the starting map (2) and the map of resolution values (3), maintaining the default values for the rest of parameters (4, 5).

Fig. 7.7 Filling in the protocol to compute the sharpened map.

After two iterations, the sharpening algorithm reached the convergence criterion, i.e. a difference between two successive iterations lower than 1%, and stopped. The two maps obtained in the respective iterations can be observed with ShowJ by clicking the black arrow shown in Fig. 7.7 (7) with the right mouse botton and selecting Open with DataViewer. Resulting map for each iteration will be shown, as indicated in Fig. 7.8. Visualization in ChimeraX is also possible selecting Open in the menu option File (Fig. 7.8 (1)).

Fig. 7.8 Sharpened maps generated after two iterations.

Additionally, by clicking Analyze Results (Fig. 7.7 (6)) the sharpened map obtained after the second iteration, i.e. the last map, can be also visualized and compared with the initial one in ChimeraX (Fig. 7.9).

Fig. 7.9 LocalDeblur last iteration sharpened map (yellow surface) and input map (grey mesh) in ChimeraX.

### 7.1.2. Sharpening with DeepEMhancer

DeepEMhancer is an alternative automatic sharpening method based on deep learning , implemented in Scipion in the protocol xmipp3-deepEMhancer (Appendix DeepEMhancer Sharpening). Open this protocol (Fig. 7.10 (1)) and complete it as indicated. Since only the refined map is available, we are not going to use half maps (2). Include your map (3), the type of normalization desired (4) and the deep learning mode to use (5), in this particular case highRes due to the map high resolution.

Fig. 7.10 Filling in the protocol to generate a sharpened map with DeepEMhancer.

After executing the protocol (Fig. 7.10 (6)), we can check the results (7). ChimeraX viewer will open and show the sharpened map compared with the initial one (Fig. 7.11).

Fig. 7.11 DeepEMhancer sharpened map (yellow surface) and input map (grey mesh) in ChimeraX.

## 7.2. Comparison of maps

Realize that at this point we have generated two optimized maps derived from the initial one. Additionally, some other maps could have been obtained using other map optimization methods. A comparison among them would be interesting to consider which one(s) of them should be used as input in next steps of modeling workflow. The ideal map for tracing the atomic structure should include as many details and connections as possible and, at the same time, preserve the density areas of the initial map. In other words, we can use the best sharpened map (with higher resolution) corroborating that it does not make up new densities, absent in the starting map. Nevertheless, selecting “the best” sharpened map could be difficult sometimes, especially if the map is very big or there are some regions optimized in one of the sharpened maps and other areas optimized in the other one. In that case, you can use several maps at the same time, having all of them perfectly aligned according to the same origin of coordinates.

In the tiny example shown in this tutorial we are working with a high resolution map and there are almost no differences in resolution between the starting map and the two derived sharpened maps, although this is not usually the case in real life. In this quite uncommon case the initial unsharpened map would be enough to trace the atomic structure. However, in order to detail the method, the starting map and their two sharpened ones will be used simultaneously.

## 7.3. Extraction of the asymmetric unit map

Since smaller volumes usually include lower number of individual structural elements, making easier fitting models in maps and simplifying modeling process, the part of the map chosen to work with will always be the smaller asymmetrical subunit of the starting loaded map, also known as asymmetric unit (ASU). The size of the ASU thus depends on the symmetry order of the initial volume. The higher the symmetry order, the smaller the ASU. The atomic structure of the whole volume will be obtained straight forward by simply repetition of the ASU structure according to the symmetry order. Then, the first step to simplify the complexity of the initial volume is extracting the ASU. This task can be accomplished by using the Scipion protocol xmipp3-extract unit cell that extracts the geometrical ASU of the map (Appendix Extract unit cell).
Fig. 7.12 shows how to fill in this protocol form (1). Consider that in this particular case the protocol will be run three times, one with each map (the initial one and the two sharpened derived ones). Include each map in a protocol form parallel to that shown in Fig. 7.12 (2). Since metHgb macromolecule shows symmetry C2, we have selected cyclic symmetry (Cn) as type of symmetry (3), and 2 as symmetry order (4). The angle offset selected (5) turns -45º around the Z axis the mask used to create the ASU. The two wizards on the right (6, 7) help you to select the radii to delimit a fraction of the map comprised between the coordinate origin (inner radius 0.0) and the maximum radius (outer radius 50.0). The final extracted volume will be slightly higher than the ASU due to the expand factor 0.2 (8). The respective tutorial appendix Extract unit cell includes a comprehensive explanation of the meaning of parameters.

Fig. 7.12 Extracting the map asymmetric unit (ASU).

After executing the protocol (Fig. 7.12 (9)), the resulting expanded ASU can be observed (10) with ChimeraX (Fig. 7.13). Note the additional expanded volume of the ASU on the left side of the figure. The ASU itself, on the right side, constitutes the half volume. Since the total volume contains the structure of four proteins, we can anticipate that this smaller asymmetrical subunit of the initial volume contains two proteins, one \alpha and one \beta subunit. Then, the respective structures of these two proteins could be fitted in the map ASU simultaneously or in successive modeling workflow steps.

Fig. 7.13 Expanded ASU (yellow-green-blue) and initial volume (gray) visualized with ChimeraX. The purple broken line delimits the ASU (right) and the expanded volume (left).