Introduction
AlphaFold is an artificial intelligence based software developed by DeepMind Technologies, a subsidiary of Alphabet (parent company of Google) which performs protein structure prediction using a deep learning system. Initial demonstration of the program at CASP13 showed great success at accurately predicting structures for targets rated as being the most difficult, where no previous template existed. In a second public demonstration during the 14th CASP in 2020, the AlphaFold team achieved a level of accuracy much higher than any other participating group.
Such results have proven to be a giant step forward in the field of ab initio protein structure prediction. Later in July 2021, DeepMind published a paper in Nature describing AlphaFold. It also made public the source code of the software on GitHub and offered a searchable database containing the structure prediction result of the complete proteome of about 20 species (over 365 000 protein structures).
How AlphaFold works
Using AlphaFold via SBGrid
SBGrid is a software infrastructure that includes a library of over 400 structural biology applications, including AlphaFold. The SBGrid software collection is already installed on the bioinformatics infrastructure of the structural biology platform and can be accessed either remotely via SSH or physically by logging into one of the computers in room C-305 (Roger-Gaudry building). More details are given on the SBGrid Wiki page for AlphaFold.
...
- A modified version of the AlphaFold wrapper provided by SBGrid that can be downloaded here.
- Your protein sequence in FASTA format (a complete description of the FAST format can be found here). The description line is mandatory. An example file can be downloaded here.
...
| Avertissement |
|---|
If you do not specify the output directory, it will be saved , by default to |
It will take over 60 minutes for the calculation to take place on mazuelo. Once it is done, you can go to the results folder and analyze your results. A description of the various files produced by the program is provided on the GitHub page of AlphaFold.
Using AlphaFold via a web interface
DeepMind, in collaboration with Google, has provided users with a Colab notebook to run AlphaFold. The notebook can be opened at the address below, and Instructions are provided in the notebook as to how to proceed:
https://colab.research.google.com/github/deepmind/alphafold/blob/main/notebooks/AlphaFold.ipynb
| Avertissement |
|---|
As stated in the Colab notebook: "In comparison to AlphaFold v2.0, this Colab notebook uses no templates (homologous structures) and a selected portion of the BFD database. We have validated these changes on several thousand recent PDB structures. While accuracy will be near-identical to the full AlphaFold system on many targets, a small fraction have a large drop in accuracy due to the smaller MSA and lack of templates. For best reliability, we recommend instead using the full open source AlphaFold, or the AlphaFold Protein Structure Database." |