Implementation of paper:

Hu, F., Hu, Y., Zhang, W., Huang, H., Pan, Y., & Yin, P. (2023). A Multimodal Protein Representation Framework for Quantifying Transferability Across Biochemical Downstream Tasks. Advanced Science, 2301223. https://doi.org/10.1002/advs.202301223

The data can be downloaded from these links. If you have any question, please contact [email protected].

Pretrain dataset: https://drive.google.com/file/d/1xHUs0B9VuKviBzj-k-203p4a9vEoo1RW/view?usp=sharing Downstream dataset: https://drive.google.com/file/d/10yywJNTQ9Z30B_4uyNfQhnXQdhhdjK3W/view?usp=sharing GNN-PPI data: https://drive.google.com/file/d/1YSXNsTJo-Cdxo08cHLb6ghd6noJJ4y73/view?usp=sharing GNN-PPI pretrained embedding: https://drive.google.com/file/d/1sq2VQGAMWmWg02hqhyWju2xuiJ-oHbq0/view?usp=sharing

The pre-trained model checkpoint can be downloaded from this link. If you have any question, please contact [email protected].

https://drive.google.com/file/d/1NVxB00THWxKdTZkLM7T6xdQJM_3TFMVr/view?usp=sharing

You can download this repo and run the demo task on your computing machine.

• Pre-train model.

cd Multimodal_pretrain/
python src_v0/main.py

• Fine-tune on downstream tasks using pre-trained models (downstream tasks: stability, fluorescence, remote homology, secondary structure, pdbbind, kinase).

# For example
cd Multimodal_downstream/
python src_stability/main.py

• Fine-tune on gnn-ppi using pre-trained embedding.

cd Multimodal_downstream/GNN-PPI/
python src_v0/run.py

• Guidance for hyperparameter selection.

You can select the hyperparameters of the Performer encoder based on your data and task in: