kexinhuang12345 / DeepPurpose

A Deep Learning Toolkit for DTI, Drug Property, PPI, DDI, Protein Function Prediction (Bioinformatics)
https://doi.org/10.1093/bioinformatics/btaa1005
BSD 3-Clause "New" or "Revised" License
970 stars 272 forks source link

Do you include MolTrans in this repo #55

Closed xuzhang5788 closed 3 years ago

xuzhang5788 commented 3 years ago

You have another DTI repo called MolTrans. I think you didn't include this model in this toolkit, am I right? If not, what is the difference between DTI model of MolTrans and other models in this repo. Which one is better? Thanks a lot.

kexinhuang12345 commented 3 years ago

Hi, currently, it is not included. MolTrans should have a better predictive performance for DTI but it has more complicated model architectures and also takes lots of time to train and tune hyperparameters. DeepPurpose is designed for fast prototyping and also supports Drug property, protein function, DDI, PPI predictions. It is for direct downstream use. The model architectures are also very simple (encoder-decoder framework).

kexinhuang12345 commented 3 years ago

So, depends on your use case. If it is for industry use, I would recommend DeepPurpose. If it is for research benchmarking, you could include both DeepPurpose and MolTrans.

xuzhang5788 commented 3 years ago

Thank you

xuzhang5788 commented 3 years ago

I found you still have an option of embedding method as transformer both for drug and protein. Are they different from MolTrans? Can I choose the transformer option for the drug or protein embedding method? Thanks

kexinhuang12345 commented 3 years ago

Hi, they are different. Since MolTrans has a specific decoder with interaction map + CNN.

xuzhang5788 commented 3 years ago

Thank you. Actually, I tried your MolTrans, but I got errors. If you have time, please visit MolTrans repo. Hope to get your response. Thanks a lot.

kexinhuang12345 commented 3 years ago

Oh thanks for the reminder. somehow i didn't receive notification for that repo. I will checkout in the next two days.