# ISM

By [Jeffrey Ouyang-Zhang](https://jozhang97.github.io/), [Chengyue Gong](https://sites.google.com/view/chengyue-gong), [Yue Zhao](https://zhaoyue-zephyrus.github.io), [Philipp Krähenbühl](http://www.philkr.net/), [Adam Klivans](https://www.cs.utexas.edu/users/klivans/), [Daniel J. Diaz](http://danny305.github.io)

This repository contains the model presented in the paper [Distilling Structural Representations into Protein Sequence Models](https://www.biorxiv.org/content/10.1101/2024.11.08.622579v1).
The official github can be found at https://github.com/jozhang97/ism.

**TL; DR.** ESM2 with enriched structural representations

## Quickstart

This quickstart assumes that the user is already working with ESM-C and is interested in replacing ESM-C with ISM-C. First, download ISM-C.
```bash
# recommended
huggingface-cli download jozhang97/ismc-600m-2024-12 --local-dir /path/to/save/ismc

# alternative
git clone https://huggingface.co/jozhang97/ismc-600m-2024-12
```

Add the following lines of code.
```
from esm.models.esmc import ESMC
model = ESMC.from_pretrained("esmc_600m").to("cuda")
state_dict = torch.load('/path/to/ismc_600m_2024_12_v0.pth')
model.load_state_dict(state_dict)
```