Five ways to run AlphaFold

What is AlphaFold

  • AlphaFold can accurately predict 3D models of protein structures by providing an amino acid sequence. (Ref: Deep Mind.)
  • The AlphaFold network directly predicts the 3D coordinates of all heavy atoms for a given protein using the primary amino acid sequence and aligned sequences of homologues as inputs. (Jumper et al., 2021)
  • See also previous posts (bottom of the page.)
Predicted quaternary structure
Predicted quaternary structure (no heme or iron.)

How to access AlphaFold

There are multiple options to access AlphaFold. Here are four options using the limited databases. One options for the full databases.

Using limited Databases:

1.) Casual users can use “Google Colabwith a web browser. Familiarity with Jupyter Notebook for running Python is helpful. The service is free for up to 2hours of computations. For longer computations a Pro account might help. However, for very large structures, one of the other options may be needed. This method uses the “limited databases.”

2.) UW-Madison Biochemistry personnel can access the Biochemistry Compute Cluster (BCC) to run an implementation of the Google Colab option on the local BCC cluster. Unlike the web-browser based version, this implementation is run within a Docker environment. See upcoming separate post for details. This method uses the “limited databases.”

3.) Installing the Google Colab version on your laptop. See: AlphaFold2 on Macintosh M1

4.) Running the Google Colab version from within ChimeraX.

Using Full databases:

5.) The UW-Madison Center for High Throughput Computing (CHTC) offers free computing accounts for all. UW-Madison personel. This method requires a different container technology (Appatainer instead of Docker) and takes advantage of the complete structure, sequence, and multiple sequence alignments databases installed locally. The databases occupy about 2.5 TeraBytes of disk space. See upcoming separate post for details.

Some labs may also have their own, local cluster installations.

Getting help

I have successfully run all three methods to predict protein structures. Larger structures, including homo- or hetero- multimers require the computing power of CHTC.

If you are interested in running AlphaFold predictions you can contact me at


Jumper, J., Evans, R., Pritzel, A. et al. Highly accurate protein structure prediction with AlphaFold. Nature 596, 583–589 (2021).

Previous posts on AlphaFold:

AlphaFold2 on Macintosh M1
AlphaFold – How to interpret structures
AlphaFold – run from ChimeraX
AlphaFold code
AlphaFold background