This directory contains all code used for the paper.
The subdirectories simulated, pretrain, and function correspond to the three experiment sections in the paper.
Model weights can be found on Zenodo and the function model predictions are included in this repository which can be used to reproduce the results.
The training scripts for each model are:
- Simulated experiments:
simulated/sim_experiments.py - Unconditional pretraining:
pretrain/train.py - ESM-conditioned AF-DB pretraining:
pretrain/train_swissprot.py - Function prediction:
function/train_[mlp/transformer].py