AlphaFold is an open source implementation of the inference pipeline for AlphaFold v2, a deep learning system for protein structure prediction developed by Google DeepMind. The repository provides Python code that enables researchers and practitioners to predict three-dimensional protein structures from amino acid sequences, addressing a fundamental problem in structural biology and computational biology with applications to drug discovery and molecular modeling.
The repository includes multiple components beyond the core monomer prediction system. It provides an implementation of AlphaFold-Multimer, which predicts structures of protein complexes, though this is noted as a work in progress and less stable than the monomer system. The codebase also includes technical documentation for AlphaFold v2.3.0 describing the models and inference procedures, as well as baseline predictions from CASP15 along with documentation of manual interventions performed during those predictions.
Installation and execution require substantial computational resources and setup. The system runs on Linux only and requires an NVIDIA GPU with sufficient memory for larger protein structures. Full installation demands up to 3 TB of disk space to store genetic databases, with SSD storage recommended for performance. The setup process involves installing Docker, the NVIDIA Container Toolkit for GPU support, and downloading genetic databases including BFD, MGnify, PDB70, PDB structures, UniRef30, UniProt, and UniRef90. The download script handles acquiring these databases, which total approximately 556 GB in download size and 2.62 TB when unzipped. Users can alternatively download reduced databases for lower resource requirements.
The model parameters are available separately and include five standard models validated during CASP14, five pTM models fine-tuned to produce predicted TM-score and predicted aligned error values, and five AlphaFold-Multimer models. While the code is licensed under Apache 2.0, the model parameters and CASP15 prediction data are made available under CC BY 4.0 licensing.
The most frequently labeled issues involve third-party tools, setup procedures, and error reports. The repository's contributor network overlaps with major projects including microsoft/vscode, microsoft/typescript, and rust-lang/rust, indicating cross-pollination with significant open source ecosystems.
The repository is classified across multiple domains including protein structure prediction, deep learning, bioinformatics, protein folding, structural biology, amino acid sequences, computational biology, molecular modeling, biomolecules, and drug discovery. Users are directed to cite the original AlphaFold paper and the AlphaFold-Multimer paper in publications disclosing findings from the code or model parameters, and the supplementary information provides detailed method descriptions.
