Inquiry regarding Pre-trained Model Weights and Inference Scripts

[xinyuanzhu]

Hi there,

First of all, congratulations on this excellent work published on NMI! I've been reading through the paper/code and find the approach taken in ImmunoStruct to be very impressive and promising for structural modeling.

I am very interested in trying out the model, but I noticed that the pre-trained model weights and specific inference scripts (or instructions on how to run inference with pre-trained weights) do not seem to be available in the repository yet.

Would it be possible for you to release the checkpoints and a brief guide on how to run inference? This would be incredibly helpful for the community to reproduce the results and further explore the tool.

Thank you very much for your time and contribution!

[ChenLiu-1996]

Thank you for the interest!

Since this is a project spanning over 3 years, we have lost track of the raw PDB structure data by now. To ensure a better experience for people like you, we are recently working on rerunning the protein folding (to create the PDB files) and pytorch geometric conversion (to create the PyG files). I just successfully ran the first folding a few minutes ago, and it should take around a week. Once we validate the data and make sure the pipeline is bug-free, we will:

1. Update the data on GitHub
2. Upload the graph PDB and PyG data to huggingface.
3. Upload the pre-trained model weights to huggingface.

Later, we will also create a demo script (perhaps in Jupyter notebook) to show how to run the model on a single new sequence.