In 2018, Google’s DeepMind participated in CASP13 with their algorithm AlphaFold which dramatically improved non-templated protein prediction and its accuracy. The second version released in 2020, with improved neural networks, was able to surpass experimental data for some candidate proteins.

AlphaFold’s true worth came from its success in free modeling-predicting structures for proteins without any comparable templates, suggesting that we could now predict any protein’s structure with its primary amino acid sequence.

AlphaFold is the result of multiple iterations of the CASP structure prediction experiment and numerous experimentally derived protein structures in the Protein Data Bank. With more tweaks to its algorithm, it can now model proteins with no templates available at atomic resolution.