DFrolova/MULAN_datasets

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.

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.