[Proteopedia] AlphaFold2 continues to be the best

[Eric Martz]

Every 2 years, there is an international competition (CASP) to see which 
method can best predict protein structure from sequence. Predictions are 
made for proteins whose empirically-determined structures will be 
published later, but the predictors don't have access to those 
structures when the predictions are made. In 2020 at CASP 14, DeepMind's 
AlphaFold2 blew everyone away by predicting structures for most proteins 
with accuracies nearly as good as empirical determinations.

The results of the 2022 CASP 15 have been coming out in publications 
over recent months. AlphaFold2 continues to out-perform all other 
methods for most sequences. There are now >200 million 
AlphaFold2-predicted structures in its free database 
(http://alphafold.ebi.ac.uk). (There are roughly 50,000 
sequence-distinct proteins among the ~200,000 empirical structures in 
the Protein Data Bank.) If your protein of interest is not in the 
database, instructions on how to use free AlphaFold Colab to make 
predictions are in Proteopedia:

https://proteopedia.org/w/How_to_predict_structures_with_AlphaFold

A one-paragraph summary of the results of CASP 15 is now available:

https://proteopedia.org/w/Theoretical_models#2022:_CASP_15

See also

https://proteopedia.org/w/AlphaFold

-Eric Martz