---
license: mit
---

This is a dataset card containing preprocessed datasets used for the evaluation of protein language models presented in the paper [MULAN: Multimodal Protein Language Model for
Sequence and Structure Encoding](https://www.biorxiv.org/content/10.1101/2024.05.30.596565v1).

Fistly, it contains the preprocessed AF-0.5M dataset used for training of MULAN-small (AF05_pretraining.zip).

Secondly, it contains zip files with all downstream datasets used in the paper. 
Each downstream dataset folder contains `id2label.json` file with targets for each dataset split and a folder `dataset` 
with the preprocessed dataset ready to pass into the `ProteinDataset` class.

You should manually download each downstream task archive as well as training data archive and use these datasets with 
the original code of the paper in [MULAN repository](https://github.com/DFrolova/MULAN).