Hello DeepChem Community 
I have been working on a project using DeepChem to develop graph neural networks for predicting protein ligand binding affinities; I have made some progress & I am facing some challenges.
Develop a GNN model to predict binding affinities of protein ligand pairs.
I am Using a combination of publicly available datasets & custom curated data.
DeepChem for model building and training, along with standard libraries like TensorFlow & PyTorch.
My model may overfit on the training data even with dropout and other regularization strategies applied. Any recommendations for improving the model’s generalization?
I am using atom and bond features as input to the GNN. There are any features or
preprocessing steps that could enhance model performance?
Has anyone used transfer learning to solve comparable issues successfully? If yes, whatever
pretrained models or methods would you suggest?
For more information I have check this- https://forum.deepchem.io/t/using-a-preprocessed-version-of-qm9-with-deepchems-modelsmulesoft-api-integration
Thank you in advance for your support and assistance. 