AlphaFold2 is a deep learning algorithm that leverages techniques originating in natural language processing to do protein structure prediction. It was announced by DeepMind in 2020 at the CASP 14 competition where it stunned the competition with its performance. The creators have given several talks and presentations on the algorithm (see, e.g., here), the model and codebase has been very recently (Jul 15, 2021) open sourced (under Apache License, Version 2.0 and under Commons Attribution-NonCommercial 4.0 International respectively). This project is intented at creating a slim and fast codebase and model weights under some unrestrictive license (MIT License, Apache 2.0 or similar).

As of , our goal until now has been to attempt a replication of he model's performance given the limited knowledge we had (codebase was not yet open sourced). The main scope should be shifted towards replicating the existing codebase and providing free model weights now that the architecture is available.
As of , the main components of the model are being coded and assembled, building on top of previous months' work.
Next steps:
Our aim is to build a modern, modular and extensible codebase, which could power the next round of improvements over the current architecture, as well as being efficient to reduce the computational cost of predicting many protein structures. We are also planning to provide multiple intermediate represenattions (ie. embeddings) which could help develop explainability and downstream applications.