Protein property prediction code

[Tian-science]

Nice job you have done! I noticed in your article that you have done some work on protein property prediction. Could you please upload the code for this kind of task, or give me some advice on how to modify the code?

[kxz18]

Hi, it should be similar to graph classifier if the property is discrete. Or affinity predictor if it is continuous. The protein property prediction problem in the paper further utilizes IEConv layers for fair comparison with the baselines, which is not the contribution in our model, therefore I chose to not include it in this repo.

[Tian-science]

Thank you! Could you please give me some advice on how to pre-process the single protein data (the embedding process),
and what the parameters in config file (like the xx.json one in scripts/exps/configs/xx/), so I can do some experiments on it?

[kxz18]

For the data processing, it should be similar to the processing of PDBbind, except that there is not small molecule ligand here. But you need to adjust the saved information (i.e. affinity) to the property score according to your needs. As for the config, it should be like the PPA config. Further, you need to manually specify the type of dataset (either using existing ones or defining your own class) in train.py based on the task name. However, if you are already working on some projects for protein property prediction, I would advice integrating the cleaned model into your coding framework, which might be easier to develop.

[Tian-science]

Thank you!