search

samsledje / ConPLex

Adapting protein language models and contrastive learning for highly-accurate drug-target interaction prediction.
http://conplex.csail.mit.edu
MIT License
119 stars 32 forks source link

Fine-tuning ConPLex on custom datasets #26

Closed seyonechithrananda closed 1 year ago

seyonechithrananda commented 1 year ago

Is there any plans to support custom dataset contrastive fine-tuning? And if not, do you have any advice on good first steps to do so ourselves?

samsledje commented 1 year ago

Hi @seyonechithrananda ,

Custom training data for both binary and contrastive training is on our list, but is currently lower priority as we're currently focused more on making the inference mode with pre-trained models available. To get started yourself, I'd recommend looking at the ContrastiveDataset and DUDEDataModule classes. These are a good place to get started for defining your own contrastive fine-tuning data sets, as they should have the same functionality as this DataModule, which is created here in training.