DeepChem-Compatible RFDiffusion Implementation

This project aims to implement state-of-the-art protein design models, RFDiffusion and RFDiffusion-2, within the DeepChem framework using a clean, end-to-end PyTorch pipeline. The goal is to make advanced generative protein modeling accessible through standardized DeepChem abstractions such as TorchModel, Dataset, and Featurizer, facilitating integration with existing deepchem workflows. The model leverages diffusion-based generative techniques to transform random noise into realistic protein structures, built on a RoseTTAFold inspired architecture that jointly captures sequence, pairwise interactions, and 3D geometry. By incorporating SE(3)-equivariant operations, the system ensures physically consistent predictions while maintaining modularity through the use of existing equivariant libraries.