
by University of California San Diego
CHGNet is a pretrained universal neural network potential for charge-informed atomistic modeling. It predicts energy, force, stress and magnetic moments of crystal structures at near DFT accuracy.
Web appFree
Step 1: Install the package with pip install chgnet.
Step 2: Load the pretrained model with CHGNet.load().
Step 3: Read a structure with pymatgen and call predict_structure.
Step 4: Run relaxation or molecular dynamics with the included classes.
Step 5: Fine tune the model on your system using the tuning notebook.
Energy in eV per atom, forces, stress in GPa and magnetic moments for a given structure.
A default model trained on Materials Project trajectories, a deprecated earlier version, and an r2scan-level model. Newer MatPES-based checkpoints live in the MatGL repo.
CPU, NVIDIA GPUs, Intel GPUs and Apple silicon. The device is picked automatically and can be overridden per call or with an environment variable.
Yes. The Tuning CHGNet notebook shows how to fine tune the pretrained model to your system.
Boltz-2Biomolecular AI model extending structure prediction toward molecular interaction and binding-affinity modeling
DeepCellDeep-learning platform and models for cell segmentation, tracking, and quantitative analysis of microscopy images
DeepImageJDeepImageJ is a plugin that lets you run pre-trained deep learning models inside ImageJ and Fiji. It needs no deep learning expertise or programming skills.