Models

This ESM entry covers the Meta FAIR repository, including ESM-2 protein representations and ESMFold structure prediction.

ProteinMPNN designs amino-acid sequences from protein backbones and provides full-backbone and Cα-only models.

RFdiffusion is a protein backbone diffusion project supporting unconditional generation, motif constraints and binder design.

Molecular Generation

OpenFold is a trainable PyTorch implementation of AlphaFold 2 for protein structure inference and training.

Boltz is a biomolecular interaction model family. Boltz-2 provides workflows for complex structure and binding affinity prediction.

Drug Discovery

SevenNet provides graph neural network interatomic potentials, pretrained models, fine-tuning interfaces and ASE/LAMMPS integration.

Computational Chemistry

MS2DeepScore predicts molecular structural similarity from pairs of tandem mass spectra using a Siamese neural network, with tools for spectral embeddings, inference and custom model training.

Spectroscopy

Spec2Vec is a Python package for MS/MS spectral similarity scoring, using Word2Vec embeddings learned from mass fragments and neutral losses to compare query and reference spectra.

Spectroscopy

DeePKS-kit is a Python toolkit for training quantum-chemistry energy functionals, testing post-HF models, and running self-consistent calculations through the DeePHF and DeePKS schemes.

Quantum Chemistry

DeePTB is a Python package for deep-learning electronic-structure models, combining environment-dependent tight binding with equivariant Hamiltonian, density-matrix and overlap-matrix representations.

Quantum Chemistry · Materials Discovery

CGCNN implements crystal graph convolutional neural networks for learning material properties from crystal structures, with custom-data training and prediction using pre-trained models.

Materials Discovery

CrabNet implements an attention-based model for predicting material properties from composition alone, with companion guidance on basic use and model interpretability.

Materials Discovery

ALIGNN provides atomistic graph neural networks for materials property prediction, with training workflows, pretrained predictors and ALIGNN-FF force fields for structural optimization.

Materials Discovery

M3GNet is an archived materials graph neural network implementation with three-body interactions, a pretrained interatomic potential, and workflows for crystal relaxation, molecular dynamics and model training.

Computational Chemistry · Materials Discovery

CHGNet is a pretrained, charge-informed neural network potential for crystal structures, predicting energies, forces, stresses and magnetic moments for relaxation and molecular dynamics workflows.

Computational Chemistry · Materials Discovery

MEGNet is a deprecated TensorFlow graph-network implementation with pretrained molecular and crystal property models, training utilities, and transferable elemental embeddings.

Molecular Property Prediction · Materials Discovery

MODNet is a Python framework for supervised materials-property prediction from composition or crystal structure, with joint learning for multiple targets and two documented pretrained models.

Materials Discovery

Catalysis-oriented atomistic modeling workflows in fairchem, using pretrained UMA models and ASE calculators for surface relaxation, energy prediction and molecular dynamics.

Materials Discovery

REINVENT is a Python molecular-design toolkit that uses generative models, reinforcement learning and transfer learning for configurable small-molecule generation and optimization.

Molecular Generation · Drug Discovery

RetroXpert is a two-stage retrosynthesis research implementation that predicts product bond disconnections, then generates reactants from synthons using an OpenNMT-based model.

Retrosynthesis