Alphafold (github.com)

Article summary

The AlphaFold package provides an implementation of the inference pipeline of AlphaFold v2, a model for predicting protein structures. The package includes a Docker script for easy installation and running, and requires a machine with a Linux operating system, a modern NVIDIA GPU, and significant disk space. The model parameters and genetic databases are available for download, with the total download size being around 556 GB and the total size when unzipped being 2.62 TB. The package is licensed under the Apache 2.0 License, and the AlphaFold parameters and CASP15 prediction data are made available under the terms of the CC BY 4.0 license.

Main themes

• AlphaFold
• protein structure prediction
• computational resources
• scientific collaboration
• drug discovery
• biochemistry
• computer science
• open-source software

What commenters say

• The computational resources required to run AlphaFold are significant, but not prohibitively expensive for commercial use.
• The release of AlphaFold's code and data is a significant step forward for science, allowing for easier reproduction and modification of the work.
• The model's architecture is not particularly novel, and its success may be due in part to the resources available to the developers.
• The model is not yet able to accurately predict the structures of multi-complex proteins, which are important for many biological processes and drug design.
• The critical path to drug discovery and approval does not rely heavily on protein structure determination.
• The value of AlphaFold to the pharmaceutical industry is uncertain, and may be limited by the complexity of the drug development process.
• The release of AlphaFold's code and data is a significant step forward for the scientific community, but the model's limitations and potential applications are still being debated.
• The use of AlphaFold requires significant expertise in biochemistry and computer science, and may not be accessible to all researchers.