About
ModelAngelo is an open-source automatic atomic model building program purpose-built for structural biologists working with cryo-EM data. Traditionally, fitting atomic coordinates into electron density maps is a highly manual, time-intensive process requiring expert knowledge. ModelAngelo automates this step using a combination of deep learning and protein language models, intelligently placing and tracing amino acid chains within cryo-EM density maps with minimal user input. The tool leverages GPU-accelerated inference for high-throughput processing and performs best on NVIDIA GPUs (RTX 2080 or newer) with at least 8GB of VRAM. Pre-trained model weights—totaling approximately 10GB—are downloaded at setup time. ModelAngelo can be installed on personal workstations via Conda or deployed on high-performance computing (HPC) clusters using Docker or Singularity containers, making it flexible for diverse research environments. ModelAngelo integrates naturally into existing cryo-EM pipelines alongside tools like RELION and cryoSPARC, accepting standard map formats as input and producing atomic coordinate files suitable for downstream refinement. It is particularly valuable for high-resolution maps where automated tracing can provide a reliable starting model. Hosted on GitHub under the MIT license, ModelAngelo is freely available to the research community and under active development. It is primarily aimed at structural biologists, bioinformaticians, and computational scientists involved in protein structure determination, structural genomics, and structure-based drug discovery.
Key Features
- Fully Automated Model Building: Automatically fits and traces atomic coordinates into cryo-EM density maps, eliminating the need for extensive manual intervention.
- Deep Learning & Language Model Integration: Combines convolutional deep learning with protein language models to achieve high accuracy in residue placement and chain tracing.
- GPU-Accelerated Inference: Leverages NVIDIA GPUs (RTX 2080 or newer, 8GB+ VRAM) for fast and efficient computation on large cryo-EM datasets.
- HPC Cluster & Container Support: Deployable on computational clusters via Docker and Singularity container images, enabling scalable use in institutional environments.
- Conda-Based Installation: Simple installation via Anaconda/Miniconda for personal workstations, with a provided install script for streamlined setup.
Use Cases
- Automated protein structure determination from newly resolved cryo-EM density maps
- Accelerating structural biology research pipelines by generating high-quality initial atomic models
- High-throughput structural genomics projects requiring rapid model building across many targets
- Providing starting models for downstream refinement in drug discovery and structure-based design workflows
- Academic research and teaching in structural biology, bioinformatics, and computational biochemistry
Pros
- Fully Open Source: Released under the MIT license with source code freely available on GitHub, allowing unrestricted academic and commercial use.
- Dramatically Reduces Manual Work: Automates one of the most labor-intensive steps in cryo-EM structure determination, saving researchers significant time and effort.
- Flexible Deployment: Supports personal workstations and HPC clusters through Conda, Docker, and Singularity, accommodating diverse research computing environments.
Cons
- High GPU Requirements: Requires an NVIDIA GPU with at least 8GB of VRAM, making it inaccessible to users without suitable hardware.
- Large Disk Space Needed: Pre-trained model weights for ModelAngelo and its language model total approximately 10GB, requiring significant storage allocation.
- Steep Technical Setup: Installation and usage assume familiarity with command-line tools, Conda environments, and HPC systems, posing a barrier for non-technical users.
Frequently Asked Questions
What is cryo-EM atomic model building?
Cryo-EM (cryo-electron microscopy) produces 3D density maps of proteins. Atomic model building is the process of fitting atomic coordinates—amino acid chains—into these density maps to determine the precise molecular structure. ModelAngelo automates this process using AI.
What GPU does ModelAngelo require?
ModelAngelo requires an NVIDIA GPU with at least 8GB of memory. It has been tested and performs well on GPUs from the RTX 2080 series and newer.
How do I install ModelAngelo?
For personal use, install Anaconda or Miniconda, then use the provided install_script.sh to set up the Conda environment and download the required model weights. For cluster environments, Docker and Singularity container definitions are also provided in the repository.
Is ModelAngelo free to use?
Yes. ModelAngelo is fully open-source under the MIT license, freely available on GitHub for both academic and commercial use.
What input does ModelAngelo need and what does it output?
ModelAngelo takes a cryo-EM density map as input (along with optional sequence information). It outputs atomic coordinate files suitable for downstream refinement in programs such as REFMAC, Phenix, or ISOLDE.
