Use DeepEMhancer Online

Commercially Available DeepEMhancer No-Code Web Server

DeepEMhancer: Deep Learning Post-Processing for Sharper, Cleaner Cryo-EM Maps

DeepEMhancer is a deep learning method for post-processing cryo-EM volumes that performs masking, denoising and local sharpening-like enhancement in a single step. It was developed by Ruben Sanchez-Garcia, Josue Gomez-Blanco, Ana Cuervo, Jose Maria Carazo, Carlos Oscar S. Sorzano, Javier Vargas and colleagues at the Centro Nacional de Biotecnologia-CSIC, McGill University and the Universidad Complutense de Madrid, and was published in Communications Biology (2021, open access). The code repository lists an Apache-2.0 license.

Raw cryo-EM maps are noisy, and their local resolution varies. Conventional post-processing such as global B-factor sharpening and masking can over-sharpen well-resolved regions and under-sharpen flexible ones, and methods such as LocScale need an atomic model, which is not available when you are trying to build one.

DeepEMhancer is trained to reproduce locally sharpened, masked maps and then runs on a new experimental map without an atomic model.

Property

Detail

Input

Two unprocessed half-maps (preferred) or one unprocessed full map; optional binary mask

Output

A post-processed map

Network

3D U-net-like convolutional neural network

Training data

147 EMDB maps (average 3.8 A); 107 training, 21 validation, 20 test

Model options

tightTarget (default), wideTarget, highRes

Code

github.com/rsanchezgarc/deepEMhancer

How DeepEMhancer Enhances Cryo-EM Maps

  1. Normalize: The map is resampled to 1 A per voxel and normalized using estimated noise statistics.

  2. Split into cubes: The volume is divided into overlapping 64 x 64 x 64 cubes.

  3. Apply the network: A U-net with three downsampling and three upsampling blocks predicts the enhanced cube.

  4. Reassemble: Overlapping predictions are averaged into the final map.

The network was trained against LocScale outputs, masked using masks simulated from atomic models, with maps drawn from EMDB entries better than 7 A resolution.

Results: Measured Improvement Over Input and Conventional Maps

Comparison

Median result

Correlation with LocScale targets

0.9 versus 0.6 for the input maps

FSC resolution against model-derived references

About 0.6 A improvement over input

Versus RELION postprocessing

Improvement below 0.05 A

Versus deposited, author-sharpened maps

About 0.1 A (FSC), 0.15 A (DeepRes)

DeepRes local resolution

All test cases improved, median about 0.4 A

In case studies, DeepEMhancer suppressed lipid-layer signal in an MRP1 transporter map (EMD-7099), improved beta-strand definition in Slc26a9 (EMD-4997) and improved density continuity in a SARS-CoV-2 RdRp map (EMD-30178), where the authors traced additional residues.

What is Tamarind Bio?

Tamarind Bio is a no-code bioinformatics platform built to give life scientists and researchers access to powerful computational tools. Many cutting-edge machine learning models are hard to deploy and use. Tamarind provides an intuitive, web-based environment that removes the complexity of high-performance computing, software dependencies and command-line interfaces.

The platform is designed for biologists, chemists and other researchers who may not have a background in programming or cloud infrastructure but want to run models on their own data. Key features include:

  • A user-friendly graphical interface for setting up and launching experiments

  • A robust API for integration into existing research pipelines

  • An automated system for managing and scaling computational resources

Tamarind treats information and data security as a top priority, as detailed in its Trust Center and Terms of Service.

Accelerating Discovery with DeepEMhancer on Tamarind Bio

  • Model building: Cleaner, better-connected density helps de novo tracing of residues.

  • Membrane proteins: Suppress lipid or detergent signal around a protein.

  • Heterogeneous resolution: Enhance maps whose local resolution varies across the structure.

  • Figure preparation: Produce clearer maps for visualization.

How to Use DeepEMhancer on Tamarind Bio

  1. Open the tool: Log in at tamarind.bio and open DeepEMhancer.

  2. Provide the map: Upload two unprocessed half-maps, or one unprocessed full map, from RELION or cryoSPARC.

  3. Avoid pre-processed input: The repository advises against already masked or sharpened maps; if you must use one, supply a binary mask, if exposed.

  4. Choose a model: tightTarget is the default, wideTarget suits cases that look over-masked, and highRes is suggested for maps better than 4 A, if exposed.

  5. Run: Launch the job.

  6. Download: Retrieve the post-processed map.

  7. Compare: Inspect the result next to the original map and try more than one model when possible.

  8. Use downstream: Use the map for model building or visualization, not as a replacement for the unprocessed data in refinement.

Things to Keep in Mind

  • The authors describe the method as not the ultimate solution and suggest combining post-processing techniques.

  • Validation relied on published atomic models, so users should compare against orthogonal methods.

  • The authors discourage estimating the resolution of post-processed maps.

  • Performance on uncommon post-translational modifications is limited by scarce training data, and the training set was relatively small.

  • Noise estimation can fail for hollow particles or fibers; measured noise statistics can be supplied.

Source: Sanchez-Garcia R, Gomez-Blanco J, Cuervo A, Carazo JM, Sorzano COS, Vargas J, et al. DeepEMhancer: a deep learning solution for cryo-EM volume post-processing. Communications Biology 2021; 4:874. DOI: 10.1038/s42003-021-02399-1. Code: github.com/rsanchezgarc/deepEMhancer.

Supporting 10,000+ scientists around the world,

from leading biotechs, and global biopharma