Machine Learning Interatomic Potentials (MLIP) in DeepChem using E(3)-Equivariant Neural Networks

This project aims to develop a native PyTorch-based Machine Learning Interatomic Potential (MLIP) framework within DeepChem, enabling efficient prediction of molecular energies and forces without relying on external frameworks. MLIPs approximate the potential energy surface of molecular systems, providing a scalable alternative to Density Functional Theory (DFT) for molecular simulations. The project will implement a modular, production-ready framework supporting E(3)-equivariant architectures such as NequIP and MACE, fully integrated with DeepChem’s TorchModel API, graph-based data pipelines, and equivariance utilities. Key deliverables include a generalized MLIPModel abstraction, efficient graph construction and batching pipelines, a stable joint energy–force training system, and comprehensive benchmarking on MD17 and QM9 datasets, including molecular dynamics stability validation. The final outcome will be a fully integrated and extensible MLIP module within DeepChem, enabling end-to-end molecular simulation workflows.