locscale 2.0.1

Contrast optimization for cryo-EM maps

LocScale - reference-based local sharpening of cryo-EM maps

LocScale is an automated program for local sharpening of cryo-EM maps with the aim to improve their interpretability. It utilises general properties inherent to electron scattering from biological macromolecules to restrain the sharpening filter. These can be provided either from an existing atomic model, or inferred directly from the experimental density map.

New in LocScale 2.0:

• Model-free sharpening: LocScale now supports reference-based sharpening without the need to supply an atomic model

• EMmerNet: a physics-inspired deep convolutional neural network sharpening method.

• Completely automated process for local map sharpening

• Full support for point group symmetry (helical symmetry to follow)

LocScale is distributed as a portable stand-alone installation that includes all the needed libraries from: https://gitlab.tudelft.nl/aj-lab/locscale/releases.

Please note that there is a GUI implemented version available as part of the CCP-EM project; it is also implemented in Scipion. Note that currently the GUI implementations only support an older version of Locscale (Locscale 1.0, with only model-based sharpening).

Installation

We recommend to use Conda for a local working environment. See here for more information on what Conda flavour may be the right choice for you, and here for Conda installation instructions.

Requirements

LocScale should run on any CPU system with Linux, OS X or Windows subsytem for Linux (WSL). To run LocScale efficiently in EMmerNet mode requires the availability of a GPU; it is possible to run it on CPUs but computation will be slow.

Installation instructions:

1. Create and activate a new conda environment

conda create -n locscale python=3.8 
conda activate locscale

Install fortran compiler

LocScale uses Fortran code to perform symmetry operations and requires a Fortran compiler to be present in your system. You can install gfortran from conda-forge.

conda install -c conda-forge gfortran

2. Install LocScale and dependencies using pip:

Recommended installation

We recommend using pip for installation. Use pip version 21.3 or later to ensure all packages and their version requirements are met.

pip install locscale 

Install development version

If you would like to install the latest development version of locscale, use the following command to install from the git repository.

pip install git+https://gitlab.tudelft.nl/aj-lab/locscale.git

To install the git repository in editable mode, clone the repository, navigate to the locscale directory, and run pip install -e .

Testing

To test functionality after installation, you can run LocScale unit tests using the following command:

locscale test

3. Install REFMAC5 via CCP4/CCPEM

LocScale needs a working instance of REFMAC5. If you already have CCP4/CCPEM installed, you can skip this step. Check if the path to run refmac5 is present in your environment.

which refmac5

If no valid path is returned, please install CCP4. Note that you can provide an alternative path to REFMAC5 during runtime by using the --refmac5_path REFMAC5_PATH flag in LocScale.

How to use

LocScale can generate locally sharpened cryo-EM maps either using model-based sharpening based on available atomic model(s), using model-free sharpening, or using deep neural network-based sharpening method (EMmerNet).

1. Run LocScale using an existing atomic model:

locscale run_locscale -em path/to/emmap.mrc -mc path/to/model.pdb -res 3 -v -o model_based_locscale.mrc

Here, emmap.mrc should be the unsharpened and unfiltered density map. If you wish to use the two half maps instead, use the following command:

locscale run_locscale -hm path/to/halfmap1.mrc,path/to/halfmap2.mrc -mc path/to/model.pdb -res 3 -v -o model_based_locscale.mrc

The output will be a locally sharpened map scaled according to the refined atomic B-factor distribution of the supplied atomic model.

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 run_locscale -em path/to/emmap.mrc -mc path/to/model.pdb -res 3 -v -o model_based_locscale.mrc -mpi

2. Run LocScale without atomic model:

If no atomic model is available, or if you do not want to use prior model information, you can use the model-free mode of LocScale. This mode will estimate the molecular volume using statistical thresholding and generate a pseudo-atomic model in the thresholded density map to approximate the distribution of atomic scatterers and estimate the local B-factor. It will then generate an average reference profile for local sharpening based on the experimental data and expected properties for electron scattering of biological macromolecules [2]. Usually all default parameters for pseudomodel and reference profile generation are fine, but you can change them if you deem fit.

locscale run_locscale -em path/to/emmap.mrc -res 3 -v -o model_free_locscale.mrc

Symmetry

If your map has point group symmetry, you need to specify the symmetry to force the pseudomodel generator for produce a symmetrised reference map for scaling. You can do this by specifying the required point group symmetry using the -sym/--symmetry flag, e.g. for D2:

locscale run_locscale -em path/to/emmap.mrc -res 3 -sym D2 -v -o model_free_locscale.mrc

LocScale currently supports all common point group symmetries. We are working on supporting helical symmetry, but this is not yet implemented.

For faster computation, use OpenMPI:

mpirun -np 4 locscale run_locscale -em path/to/emmap.mrc -res 3 -sym D2 -v -o model_free_locscale.mrc -mpi

For an exhaustive list of options, use:

locscale run_locscale --help

Alternatively, see here for more information. Please note that these pages are still being updated.

3. Run LocScale using EMmerNet predictions:

Instead of using model-based or model-free reference profiles, LocScale also supports local sharpening based on a physics-inspired deep neural network prediction method using our ensemble network EMmerNet that is under development. While we have done our very best to validate the network and mitigate the risk of hallucination, as for any such approach care needs to be exercised to avoid bias. We do encourage its use for model building and visualisation, but recommend to perform model refinement against the original map.

locscale run_emmernet -em path/to/emmap.mrc -v -trained_model model_based -gpus 1 -o emmernet_model_based.mrc

Currently, three different EMmerNet models are available and can be specified using the -trained_model flag as follows:

Model	Syntax
Model Based EMmerNet	-trained_model model_based
Model Free EMmerNet	-trained_model model_free
Ensemble EMmerNet	-trained_model ensemble

Additional models may become available and will be listed here.

For an exhaustive list of options, run:

locscale run_emmernet --help

Tutorial and FAQs

We are currently working on the tutorial and Wiki help. If you are still using LocScale 1.0, see the LocScale1-Wiki for usage instructions, FAQs and tutorial.

Credits

This project is using code from a number of third-party open-source projects. Projects used by LocScale are included under include/:

EMmer - Python library for electron microscopy map and model manipulations. License: 3-Clause BSD.
FDRthresholding – tool for FDR-based density thresholding. License: 3-Clause BSD.

LocScale also makes use of Refmac – coordinate refinement program for macromolecular structures. Refmac is distributed as part of CCP-EM.

References

If you found LocScale useful, please consider citing it:

• A.J. Jakobi, M. Wilmanns and C. Sachse, Model-based local density sharpening of cryo-EM maps, eLife 6: e27131 (2017).
• A. Bharadwaj and A.J. Jakobi, Electron scattering properties and their use in cryo-EM map sharpening, Faraday Discussions D2FD00078D (2022)

---

Bugs and questions

For bug reports please use the GitLab issue tracker.