AlphaFold
Open source implementation of AlphaFold 2, a deep learning system for highly accurate protein structure prediction.
What is AlphaFold?
AlphaFold is an open-source implementation of the AlphaFold 2 inference pipeline, a deep learning system that predicts the 3D structures of proteins from their amino acid sequences with high accuracy. It addresses a fundamental challenge in structural biology by enabling researchers to understand protein function and accelerate discoveries in fields like medicine and biotechnology. The package provides a containerized pipeline for predicting structures of both single protein chains (monomers) and protein complexes (multimers).
Bioinformaticians, computational biologists, and structural biology researchers who need to predict protein structures computationally, particularly those with access to Linux systems, modern NVIDIA GPUs, and substantial storage (up to 3 TB).
Developers choose AlphaFold for its state-of-the-art accuracy competitive with experimental methods, as demonstrated in CASP14, and its comprehensive open-source implementation that includes confidence metrics, template search, and relaxation steps. Its unique selling point is making breakthrough protein structure prediction accessible to the scientific community with a robust, Dockerized pipeline.
Overview
Open source code for AlphaFold 2.
Use Cases
Best For
- Predicting the 3D structure of a single protein chain (monomer) from its amino acid sequence.
- Modeling protein complexes (multimers), including homomers and heteromers, to understand interactions.
- Accelerating drug discovery and biomedical research by providing reliable protein structure predictions without experimental methods.
- Conducting structural biology research where experimental determination (e.g., X-ray crystallography) is challenging or infeasible.
- Educational or benchmarking purposes in computational biology, using a well-documented, high-accuracy model.
- Large-scale inference on multiple proteins, leveraging the pipeline's support for batch processing and precomputed MSAs.
Not Ideal For
- Researchers without access to high-performance computing (e.g., modern NVIDIA GPUs and 3+ TB of fast storage)
- Projects requiring real-time or rapid protein structure predictions, as inference can take hours to days for large proteins
- Teams working exclusively on Windows or macOS, since AlphaFold only supports Linux
- Applications focused on protein dynamics, ligand binding, or functional annotation, as AlphaFold predicts static structures without biochemical insights
Pros & Cons
Pros
Achieves accuracy competitive with experimental methods like X-ray crystallography, as validated in CASP14, making it highly reliable for structural predictions.
Provides per-residue pLDDT scores and predicted aligned errors (PAE) for assessing prediction reliability, which are stored in output files for detailed analysis.
Can predict structures for single protein chains and protein complexes (e.g., homomers, heteromers), addressing a wide range of biological questions from the README examples.
Offers a containerized setup that simplifies installation and ensures consistency, though it requires Docker and NVIDIA Container Toolkit configuration.
Generates multiple model predictions, rankings, raw data, and supports options like reduced databases or precomputed MSAs for tailored use cases.
Cons
Requires up to 3 TB of disk space for full databases (556 GB download) and a modern NVIDIA GPU, creating a high barrier to entry for many researchers.
Installation involves multi-step scripts, large database downloads, Docker builds, and GPU configuration, which the README notes can lead to opaque errors and slow builds.
Only runs on Linux, excluding users on Windows or macOS without complex workarounds like virtualization, as stated in the 'Installation' section.
The README explicitly warns that AlphaFold-Multimer is 'a work in progress' and 'not expected to be as stable' as the monomer system, affecting reliability for protein complexes.
Frequently Asked Questions
Related Projects
Bootstrap Your Own LatentThis repository contains implementations and illustrative code to accompany DeepMind publications
Adversarial RobustnessThis repository contains implementations and illustrative code to accompany DeepMind publications
Found a gem we're missing?
Open-Awesome is built by the community, for the community. Submit a project, suggest an awesome list, or help improve the catalog on GitHub.