Attempt at reproduction of AlphaFold2. This repository is an ecxample of dataset-driven model development. First, we generate a dataset, using a procedure that mimics some aspect of the real data. Then we develop a model, that tries to learn this particular dataset.
- toy_gpt : GPT model, mainly Karpathy's code but rewritten in a more structured way
- toy_se3 : Iterative SE(3)-transformer and simple particle dynamics dataset
- toy_prot: Toy protein dataset and structural part of the AlphaFold2 model
- toy_msa: Toy multiple sequence alignment dataset, supervised case and complete MSA+structural parts of AlphaFold2