Use Model Angelo Online
Commercially Available Model Angelo No-Code Web Server
ModelAngelo: Automated Atomic Model Building and Protein Identification in Cryo-EM Maps
ModelAngelo is a machine-learning method that automatically builds atomic models into cryo-EM density maps and can identify proteins of unknown sequence. It was developed by Kiarash Jamali, Sjors H. W. Scheres and colleagues, with the corresponding author at the MRC Laboratory of Molecular Biology in Cambridge, and published in Nature in 2024. The code is open source under the MIT license.
Building an atomic model into a cryo-EM map is slow, expert-driven work. Residues must be placed, sequences assigned and chains traced, often by hand. Earlier automated tools lose accuracy as local resolution worsens. ModelAngelo combines map density, the protein sequence and structural geometry in a single graph neural network to close that gap.
Property | Detail |
|---|---|
Publication | Nature 628, 450-457 (2024), DOI 10.1038/s41586-024-07215-4 |
Inputs | Cryo-EM map, optional protein or nucleotide sequences, optional proteome for HMM search |
Outputs | Atomic model, per-residue confidence in the B-factor column, HMM search hits |
Training data | 3,715 map-model pairs from EMDB and PDB |
Speed (single A100) | 2 minutes to 53 minutes across test structures |
How ModelAngelo Builds Models from Cryo-EM Density
Locate residues: a convolutional network predicts C-alpha positions for amino acids and phosphate positions for nucleotides, then links each residue to its 20 nearest neighbours.
Refine with a graph neural network: an eight-layer network with cryo-EM, sequence and invariant point attention modules refines positions and predicts residue identities and torsion angles.
Use sequence information: user sequences are embedded with the ESM-1b protein language model; nucleic acids do not use this module.
Assemble the model: predictions become an atomic model with idealized geometry, and residue probabilities feed HMMER searches.
Run sequence-free if needed: a mode without sequence input predicts probabilities for all 20 amino acids to identify unknown proteins.
Accuracy and Resolution Benchmarks for Cryo-EM Model Building
Test | Reported result |
|---|---|
177-structure protein test set | 77% of 410,585 residues built, 85% after excluding low Q-score residues |
Backbone accuracy | R.m.s.d. below 1.0 A even for residues with Q-scores as low as 0.4 |
Resolution versus DeepTracer | About 80% built to 3.5-4 A local resolution, versus 2.5-3 A for DeepTracer |
Large supercomplex (PDB 7Y5E, 16.7 MDa) | Six unidentified chains identified, 23 hours runtime |
Identifying Unknown Proteins
In Chlamydomonas axonemal maps, the authors identified four radial spoke proteins (RSP24-27) and two central apparatus proteins (FAP92, FAP374).
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 ModelAngelo on Tamarind Bio
Faster structure determination: generate a first atomic model from a map to cut manual building time.
Unknown protein identification: use the sequence-free mode with a proteome to name unassigned densities.
Large complexes: build models for megadalton assemblies such as the 16.7 MDa supercomplex in the paper.
Nucleic acid backbones: trace RNA and DNA in ribosome-like complexes, with validation.
How to Use ModelAngelo on Tamarind Bio
Log in and open the tool: sign in at tamarind.bio and select ModelAngelo.
Upload the cryo-EM map: provide your density map file.
Add sequences: supply protein or nucleotide sequences if known, or choose the sequence-free build option if exposed.
Add a proteome: for unknown chains, provide a proteome for the HMM search, if exposed.
Run: submit the job and wait for the model.
Download outputs: collect the atomic model and any HMM search results.
Check confidence: per-residue confidence is stored in the B-factor column, scaled 0 to 1.
Refine: take the model into refinement software such as Servalcat or Phenix and check it manually.
Things to Keep in Mind
Models still need refinement and human checking.
Regions worse than about 3.5-4.0 A need manual checking; there is no rigid-body fitting or loop building there.
Purines and pyrimidines are hard to tell apart unless resolution is better than about 2.5 A.
Cofactor densities are not modelled.
Sequence-free identification depends on the proteome searched.
Source: Jamali K, Kimanius D, Scheres SHW and colleagues, "Automated model building and protein identification in cryo-EM maps", Nature 628, 450-457 (2024), DOI 10.1038/s41586-024-07215-4. Code: https://github.com/3dem/model-angelo (MIT license).