Freezing weights and fine-tuning MPNN model

[Rhys-McAlister]

Hi, I would like to utilise transfer learning similar to how it is outlined in this post: https://keras.io/guides/transfer_learning/ and was wondering if there were any examples on freezing the weights of the MPNN component of the model whilst allowing the FFNN component to be trained with a different dataset as in "Predicting Infrared Spectra with Message Passing Neural Networks" paper.

Thanks