I understand there are multiple versions used in AlphaFold. What kind of deep learning model does the more advanced version use? CNN, RNN, or something else?
(Additionally, is there an open-source reference model for the protein folding problem?)
Artificial Intelligence Stack Exchange is a question and answer site for people interested in conceptual questions about life and challenges in a world where "cognitive" functions can be mimicked in purely digital environment. It only takes a minute to sign up.
Sign up to join this communityThe accurate answer is 'we don't know for sure'. But that said, the Deepmind article about AlphaFold 2, which won the CASP 14 competition, says this:
A folded protein can be thought of as a “spatial graph”, where residues are the nodes and edges connect the residues in close proximity. This graph is important for understanding the physical interactions within proteins, as well as their evolutionary history. For the latest version of AlphaFold, used at CASP14, we created an attention-based neural network system, trained end-to-end, that attempts to interpret the structure of this graph, while reasoning over the implicit graph that it’s building. It uses evolutionarily related sequences, multiple sequence alignment (MSA), and a representation of amino acid residue pairs to refine this graph.
A reasonable assumption that can be formed from this paragraph is that it is using some kind of a Transformer-based architecture. Yannic Kilcher's review video on AlphaFold discusses this assumption, along with a detailed explanation of the previous version of AlphaFold.
The source code for this previous version of AlphaFold, the one that was used in CASP 13, seems to have been open-sourced by Deepmind in their Github repository:
https://github.com/deepmind/deepmind-research/tree/master/alphafold_casp13