Feature-enhanced maps
Confidence-weighted map optimisation
LocScale also supports confidence-weighted density modification based on a Bayesian-approximate implementation of EMmerNet, whiich strives to simultaneously optimise high-resolution detail and contrast of low(er) resolution map regions or contextual stucture. To mitigate any risk of bias from network hallucination, LocScale integrates this procedure with calculation of a per-pixel confidence score that effectively highlights regions requiring cautious interpretation.
Confidence-weighted map optimisation workflow (Feature-enhanced maps)
GPUs required
Use of this option requires the availability of GPUs. It is possible to run the predictions on CPU-only setups but this will be very slow.
Usage
locscale feature_enhance -hm path/to/halfmap1.mrc path/to/halfmap2.mrc -v -gpus 1 -o feature_enhanced.mrc
Here, halfmap1.mrc and halfmap2.mrc should be the unsharpened and unfiltered half maps from yourr 3D refinement. If you wish to use the full map instead, use the following command:
locscale -em path/to/fullmap.mrc -mc path/to/model.pdb -v -gpus 1 -o feature_enhanced.mrc
Point group symmetry
If your map has point group symmetry, you need to specify the symmetry to force symmetrisation of the optimised map. You can do
this by specifying the required point group symmetry using the -sym/--symmetry flag, e.g. for D2:
locscale -hm path/to/halfmap1.mrc path/to/halfmap2.mrc -v -sym D2 -gpus 1 -o feature_enhanced.mrc
The output will be feature-enhanced map along with its confidence scores that can be found in the file pvDDT.mrc.
Recommended use of unfiltered input maps
Note that using unfiltered maps as input is essential. If using previously filtered maps, information beyond the spatial filter cutoff cannot be recovered.
Speed-up computation on multiple CPUs
To speed up computation, you can use multiple CPUs if available. LocScale uses OpenMPI/mpi4py for parallelisation, which should have been automatically set up during installation. You can run it as follows:
mpirun -np 4 locscale -hm path/to/halfmap1.mrc path/to/halfmap2.mrc -v -gpus 1 -o feature_enhanced.mrc -mpi
joblib. In this case, simply specify the number of CPU cores using the -np flag as follows:
locscale -hm path/to/halfmap1.mrc path/to/halfmap2.mrc -v -gpus 1 -np 4 -o feature_enhanced.mrc
Interpreting pVDDT scores
pVDDT scores
LocScale Feature-Enhanced Maps computes a voxel-wise confidence level of the optimised map, which we call the predicted Voxel-Wise
Difference Test (pVDDT) score. You should always inspect LocScale FEM maps together with its confidence scores as described
below.
Visualising confidence-weighted maps in ChimeraX
The best way to visualise confidence scores is using the surface colour option in ChimeraX. LocScale outputs the pVDDT scores in MRC format, withc each voxel representing the pVDDT score associated with it.
In ChimeraX, if your model #1 refers to the feature enhanced map (locscale_output.mrc) and model #2 refers to the pVDDT score map (pVDDT.mrc), use the following command to visualise the confidence score superimposed on the map surface:
color sample #1 map #2 palette -95,#0000ff:-80,#00ffff:0,#00ff00:80,#ffff00:95,#ff0000
pVDDT score interpretation
pVDDT scores provide an intutive way for objective map interpretation by highlighting regions that may require caution because. these regions display density that significantly deviates from the density in amplitude-only modified maps. Note that these scores do not necessarily mean that these regions should not be interpreted, just that their confidence is low(er).
