Installation

1. Install EvoDiff according to their instructions. They recommend using python 3.8.
2. To get clustering to work with protclust, you will need to have mmseqs2 installed in the command line. If you are using conda or micromamba, this is as simple as running conda install -c bioconda mmseqs2. You can visit the mmseqs2 GitHub for more installation options.

Data Collection

1. Begin with scripts in the preprocessing directory, starting with download_data.py. By altering the main function, you can change which EC class you want to download. If you are starting, I recommend using EC 5 since it is the smallest.
2. Next, access filter.ipynb. You should modify the pandas at the beginning to import your data, wherever you stored it. From here, the notebook will combine the data, assign it labels, and cluster it using protclust.

Training

At this point, you are ready to run. Try running train.py making sure that the data directory correctly points to your stored data. train_full.py implements several useful features like Autocast, learning rate warmup and scheduling, and advanced logging.