Open Resources

Chemistry AI models

Use these links to discover reusable model checkpoints, model papers, and implementation resources.

Open live model list Open Resources hub

Approved resources

106 approved model entries are available for search indexing and community discovery.

Model · Retrosynthesis and reaction prediction

ASKCOS retrosynthesis model system

Model-serving ecosystem for retrosynthesis, forward prediction, context recommendation, and synthesis planning. Model kind: model system. Primary use: Retrosynthesis and reaction prediction. Platform/source: ASKCOS.

Retrosynthesis and reaction predictiontrained modelchemistry modelretrosynthesisforward prediction
Model · Retrosynthesis and reaction prediction

AiZynthFinder trained policy models

Retrosynthetic planning system using trained expansion policies and stock filters. Model kind: model system. Primary use: Retrosynthesis and reaction prediction. Platform/source: GitHub.

Retrosynthesis and reaction predictiontrained modelchemistry modelretrosynthesispolicy model
Model · Protein-ligand docking and binding

Boltz biomolecular interaction models

Biomolecular interaction model family for structure and interaction prediction. Model kind: model system. Primary use: Protein-ligand docking and binding. Platform/source: GitHub.

Protein-ligand docking and bindingtrained modelchemistry modelprotein ligandbinding
Model · Protein-ligand docking and binding

BoltzGen binder design

Universal binder design model system built around Boltz-style biomolecular modeling. Model kind: model system. Primary use: Protein-ligand docking and binding. Platform/source: GitHub.

Protein-ligand docking and bindingtrained modelchemistry modelbinder designprotein ligand
Model · Generative molecular design

ChemFM/ChemFM-1B

One-billion parameter chemical foundation model for molecule representation and generation. Model kind: checkpoint. Primary use: Generative molecular design. Platform/source: Hugging Face.

Generative molecular designtrained modelchemistry modelchemical foundation modelSMILES
Model · Generative molecular design

ChemFM/ChemFM-3B

Three-billion parameter chemical foundation model trained on UniChem SMILES. Model kind: checkpoint. Primary use: Generative molecular design. Platform/source: Hugging Face.

Generative molecular designtrained modelchemistry modelchemical foundation modelSMILES
Model · Generative molecular design

ChemFM/ChemFMv2-20M

Small ChemFM v2 checkpoint for lightweight chemical foundation-model experiments. Model kind: checkpoint. Primary use: Generative molecular design. Platform/source: Hugging Face.

Generative molecular designtrained modelchemistry modelchemical foundation modelSMILES
Model · Property prediction and ADMET

ChemFM/admet_bbb_martins

ChemFM ADMET checkpoint for blood-brain-barrier prediction. Model kind: checkpoint. Primary use: Property prediction and ADMET. Platform/source: Hugging Face.

Property prediction and ADMETtrained modelchemistry modelBBBADMET
Model · Property prediction and ADMET

ChemFM/admet_cyp3a4_veith

ChemFM ADMET checkpoint for CYP3A4 inhibition prediction. Model kind: checkpoint. Primary use: Property prediction and ADMET. Platform/source: Hugging Face.

Property prediction and ADMETtrained modelchemistry modelCYP3A4metabolism
Model · Property prediction and ADMET

ChemFM/admet_dili

ChemFM ADMET checkpoint for drug-induced liver injury prediction. Model kind: checkpoint. Primary use: Property prediction and ADMET. Platform/source: Hugging Face.

Property prediction and ADMETtrained modelchemistry modelDILItoxicity
Model · Property prediction and ADMET

ChemFM/admet_herg

ChemFM ADMET checkpoint for hERG liability prediction. Model kind: checkpoint. Primary use: Property prediction and ADMET. Platform/source: Hugging Face.

Property prediction and ADMETtrained modelchemistry modelhERGADMET
Model · Property prediction and ADMET

ChemFM/admet_solubility_aqsoldb

ChemFM ADMET checkpoint for AqSolDB solubility prediction. Model kind: checkpoint. Primary use: Property prediction and ADMET. Platform/source: Hugging Face.

Property prediction and ADMETtrained modelchemistry modelsolubilityAqSolDB
Model · Structure recognition and utility models

DECIMER chemical image recognition models

Image transformer models for converting chemical structure depictions into machine-readable structures. Model kind: model system. Primary use: Structure recognition and utility models. Platform/source: GitHub.

Structure recognition and utility modelstrained modelchemistry modelstructure recognitionimage to SMILES
Model · Property prediction and ADMET

DeepChem/ChemBERTa-100M-MLM

Larger ChemBERTa 100M masked language model checkpoint for cheminformatics. Model kind: checkpoint. Primary use: Property prediction and ADMET. Platform/source: Hugging Face.

Property prediction and ADMETtrained modelchemistry modelChemBERTamasked language model
Model · Property prediction and ADMET

DeepChem/ChemBERTa-10M-MLM

Compact ChemBERTa 10M masked language model for molecular ML experiments. Model kind: checkpoint. Primary use: Property prediction and ADMET. Platform/source: Hugging Face.

Property prediction and ADMETtrained modelchemistry modelChemBERTacompact model
Model · Property prediction and ADMET

DeepChem/ChemBERTa-77M-MLM

ChemBERTa 77M masked language model for molecular representation learning. Model kind: checkpoint. Primary use: Property prediction and ADMET. Platform/source: Hugging Face.

Property prediction and ADMETtrained modelchemistry modelChemBERTamasked language model
Model · Property prediction and ADMET

DeepChem/ChemBERTa-77M-MTR

ChemBERTa 77M multi-task regression model for molecular property tasks. Model kind: checkpoint. Primary use: Property prediction and ADMET. Platform/source: Hugging Face.

Property prediction and ADMETtrained modelchemistry modelChemBERTamulti-task regression
Model · Property prediction and ADMET

DeepChem/MoLFormer-c3-1.1B

ChemBERTa-3 MoLFormer 1.1B masked language model for molecular representations. Model kind: checkpoint. Primary use: Property prediction and ADMET. Platform/source: Hugging Face.

Property prediction and ADMETtrained modelchemistry modelMoLFormerChemBERTa-3
Model · Property prediction and ADMET

DeepChem/MoLFormer-c3-550M

ChemBERTa-3 MoLFormer 550M molecular representation model. Model kind: checkpoint. Primary use: Property prediction and ADMET. Platform/source: Hugging Face.

Property prediction and ADMETtrained modelchemistry modelMoLFormerChemBERTa-3
Model · Protein-ligand docking and binding

DeepPurpose DTI models

Toolkit and trained workflows for drug-target interaction and related predictions. Model kind: model system. Primary use: Protein-ligand docking and binding. Platform/source: GitHub.

Protein-ligand docking and bindingtrained modelchemistry modelDTIdrug target
Model · Property prediction and ADMET

Derify/ChemBERTa-druglike

ChemBERTa model trained on augmented druglike molecule data. Model kind: checkpoint. Primary use: Property prediction and ADMET. Platform/source: Hugging Face.

Property prediction and ADMETtrained modelchemistry modelChemBERTadruglike molecules
Model · Protein-ligand docking and binding

DiffDock

Diffusion model for molecular docking and ligand pose prediction. Model kind: model family. Primary use: Protein-ligand docking and binding. Platform/source: GitHub.

Protein-ligand docking and bindingtrained modelchemistry modeldockingdiffusion
Model · Protein-ligand docking and binding

EquiBind

Geometric deep learning model for fast ligand binding pose prediction. Model kind: model family. Primary use: Protein-ligand docking and binding. Platform/source: GitHub.

Protein-ligand docking and bindingtrained modelchemistry modeldockinggeometric deep learning
Model · Generative molecular design

FLAG

Molecule generation for target protein binding with structural motifs. Model kind: model family. Primary use: Generative molecular design. Platform/source: GitHub.

Generative molecular designtrained modelchemistry modeltarget bindingstructural motifs
Model · Generative molecular design

FlowMol

Flow-matching model for de novo molecule generation over continuous and categorical variables. Model kind: model family. Primary use: Generative molecular design. Platform/source: GitHub.

Generative molecular designtrained modelchemistry modelflow matchingde novo design
Model · Generative molecular design

GCDM bio-diffusion

Geometry-complete diffusion generative model for 3D molecule generation and optimization. Model kind: model family. Primary use: Generative molecular design. Platform/source: GitHub.

Generative molecular designtrained modelchemistry modeldiffusion3D molecule
Model · Protein-ligand docking and binding

GNINA CNN scoring models

CNN scoring and docking model system for protein-ligand docking. Model kind: model system. Primary use: Protein-ligand docking and binding. Platform/source: GitHub.

Protein-ligand docking and bindingtrained modelchemistry modeldockingCNN scoring
Model · Generative molecular design

GT4SD/multitask-text-and-chemistry-t5-base-augm

Multitask text and chemistry T5 model for chemistry text-to-text and molecular tasks. Model kind: checkpoint. Primary use: Generative molecular design. Platform/source: Hugging Face.

Generative molecular designtrained modelchemistry modeltext chemistryT5
Model · Generative molecular design

GeoDiff

Geometric diffusion model for molecular conformation generation. Model kind: model family. Primary use: Generative molecular design. Platform/source: GitHub.

Generative molecular designtrained modelchemistry modeldiffusion3D molecule
Model · Generative molecular design

GeoLDM

Geometric latent diffusion model for 3D molecule generation. Model kind: model family. Primary use: Generative molecular design. Platform/source: GitHub.

Generative molecular designtrained modelchemistry modeldiffusion3D molecule
Model · Retrosynthesis and reaction prediction

Graph Logic Network retrosynthesis model

Graph Logic Network model for retrosynthetic planning. Model kind: model family. Primary use: Retrosynthesis and reaction prediction. Platform/source: GitHub.

Retrosynthesis and reaction predictiontrained modelchemistry modelretrosynthesisgraph model
Model · Generative molecular design

Graph-DiT

Graph diffusion transformer for multi-conditional molecular generation. Model kind: model family. Primary use: Generative molecular design. Platform/source: GitHub.

Generative molecular designtrained modelchemistry modelgraph diffusiontransformer
Model · Generative molecular design

Hierarchical Graph-to-Graph

Hierarchical molecular graph generation using structural motifs. Model kind: model family. Primary use: Generative molecular design. Platform/source: GitHub.

Generative molecular designtrained modelchemistry modelgraph generationstructural motifs
Model · Property prediction and ADMET

IBM BioMed BACE finetune

MoleculeNet BACE finetune of IBM small-molecule multiview model. Model kind: checkpoint. Primary use: Property prediction and ADMET. Platform/source: Hugging Face.

Property prediction and ADMETtrained modelchemistry modelBACEbioactivity
Model · Property prediction and ADMET

IBM BioMed BBBP finetune

MoleculeNet BBBP finetune for blood-brain-barrier prediction. Model kind: checkpoint. Primary use: Property prediction and ADMET. Platform/source: Hugging Face.

Property prediction and ADMETtrained modelchemistry modelBBBPADMET
Model · Property prediction and ADMET

IBM BioMed ClinTox finetune

MoleculeNet ClinTox finetune for clinical toxicity prediction. Model kind: checkpoint. Primary use: Property prediction and ADMET. Platform/source: Hugging Face.

Property prediction and ADMETtrained modelchemistry modelClinToxtoxicity
Model · Property prediction and ADMET

IBM BioMed ESOL finetune

MoleculeNet ESOL finetune for aqueous solubility prediction. Model kind: checkpoint. Primary use: Property prediction and ADMET. Platform/source: Hugging Face.

Property prediction and ADMETtrained modelchemistry modelESOLsolubility
Model · Property prediction and ADMET

IBM BioMed FreeSolv finetune

MoleculeNet FreeSolv finetune for hydration free energy prediction. Model kind: checkpoint. Primary use: Property prediction and ADMET. Platform/source: Hugging Face.

Property prediction and ADMETtrained modelchemistry modelFreeSolvsolvation
Model · Property prediction and ADMET

IBM BioMed HIV finetune

MoleculeNet HIV activity finetune for antiviral screening tasks. Model kind: checkpoint. Primary use: Property prediction and ADMET. Platform/source: Hugging Face.

Property prediction and ADMETtrained modelchemistry modelHIVbioactivity
Model · Property prediction and ADMET

IBM BioMed Lipophilicity finetune

MoleculeNet lipophilicity finetune of IBM small-molecule multiview model. Model kind: checkpoint. Primary use: Property prediction and ADMET. Platform/source: Hugging Face.

Property prediction and ADMETtrained modelchemistry modellipophilicityADMET
Model · Property prediction and ADMET

IBM BioMed MA-TED BBBP

IBM MA-TED small molecule finetune for BBBP prediction. Model kind: checkpoint. Primary use: Property prediction and ADMET. Platform/source: Hugging Face.

Property prediction and ADMETtrained modelchemistry modelBBBPMA-TED
Model · Property prediction and ADMET

IBM BioMed MUV finetune

MoleculeNet MUV finetune for unbiased virtual screening benchmarks. Model kind: checkpoint. Primary use: Property prediction and ADMET. Platform/source: Hugging Face.

Property prediction and ADMETtrained modelchemistry modelMUVvirtual screening
Model · Property prediction and ADMET

IBM BioMed QM7 finetune

MoleculeNet QM7 finetune for quantum-property prediction. Model kind: checkpoint. Primary use: Property prediction and ADMET. Platform/source: Hugging Face.

Property prediction and ADMETtrained modelchemistry modelQM7quantum property
Model · Property prediction and ADMET

IBM BioMed SIDER finetune

MoleculeNet SIDER finetune for side-effect prediction. Model kind: checkpoint. Primary use: Property prediction and ADMET. Platform/source: Hugging Face.

Property prediction and ADMETtrained modelchemistry modelSIDERside effects
Model · Property prediction and ADMET

IBM BioMed Tox21 finetune

MoleculeNet Tox21 finetune for toxicity classification. Model kind: checkpoint. Primary use: Property prediction and ADMET. Platform/source: Hugging Face.

Property prediction and ADMETtrained modelchemistry modelTox21toxicity
Model · Property prediction and ADMET

IBM BioMed ToxCast finetune

MoleculeNet ToxCast finetune for high-throughput toxicity assays. Model kind: checkpoint. Primary use: Property prediction and ADMET. Platform/source: Hugging Face.

Property prediction and ADMETtrained modelchemistry modelToxCasttoxicity
Model · Retrosynthesis and reaction prediction

IBM RXN reaction prediction models

Production reaction prediction, retrosynthesis, and synthesis planning model service. Model kind: model system. Primary use: Retrosynthesis and reaction prediction. Platform/source: IBM RXN.

Retrosynthesis and reaction predictiontrained modelchemistry modelreaction predictionretrosynthesis
Model · Generative molecular design

IRDiff

Interaction-based retrieval-augmented diffusion model for protein-specific 3D molecule generation. Model kind: model family. Primary use: Generative molecular design. Platform/source: GitHub.

Generative molecular designtrained modelchemistry modeldiffusionretrieval augmented
Model · Generative molecular design

Junction Tree VAE

Junction Tree Variational Autoencoder model for molecular graph generation. Model kind: model family. Primary use: Generative molecular design. Platform/source: GitHub.

Generative molecular designtrained modelchemistry modelgraph generationVAE
Model · Generative molecular design

LiGAN

Deep generative 3D grid model for structure-based drug discovery. Model kind: model family. Primary use: Generative molecular design. Platform/source: GitHub.

Generative molecular designtrained modelchemistry model3D generative modelprotein ligand
Model · Retrosynthesis and reaction prediction

LocalRetro retrosynthesis model

Template-based local reaction center retrosynthesis model. Model kind: model family. Primary use: Retrosynthesis and reaction prediction. Platform/source: GitHub.

Retrosynthesis and reaction predictiontrained modelchemistry modelretrosynthesistemplate model
Model · Generative molecular design

MOSES generative baselines

Benchmark platform and baseline trained models for molecular set generation. Model kind: model family. Primary use: Generative molecular design. Platform/source: GitHub.

Generative molecular designtrained modelchemistry modelMOSESbenchmark
Model · Generative molecular design

Microsoft MoLeR

Molecular graph generative model supporting scaffold-constrained molecule generation. Model kind: model family. Primary use: Generative molecular design. Platform/source: GitHub.

Generative molecular designtrained modelchemistry modelMoLeRscaffold constrained
Model · Generative molecular design

MolCRAFT

MolCRAFT series models for molecular generation. Model kind: model family. Primary use: Generative molecular design. Platform/source: GitHub.

Generative molecular designtrained modelchemistry modelmolecular generationgenerative design
Model · Structure recognition and utility models

MolScribe image-to-graph model

Robust molecular structure recognition model converting chemical images to molecular graphs. Model kind: model system. Primary use: Structure recognition and utility models. Platform/source: GitHub.

Structure recognition and utility modelstrained modelchemistry modelstructure recognitionimage to graph
Model · Retrosynthesis and reaction prediction

Molecular Transformer

Transformer model family for reaction prediction using molecular reaction SMILES. Model kind: model family. Primary use: Retrosynthesis and reaction prediction. Platform/source: GitHub.

Retrosynthesis and reaction predictiontrained modelchemistry modelreaction predictiontransformer
Model · Generative molecular design

NExT-Mol

Model combining 3D diffusion and 1D language modeling for 3D molecule generation. Model kind: model family. Primary use: Generative molecular design. Platform/source: GitHub.

Generative molecular designtrained modelchemistry model3D diffusionlanguage model
Model · Generative molecular design

NVIDIA BioNeMo GenMol

Masked discrete diffusion and fragment-based molecule generation model using SAFE molecular representation. Model kind: checkpoint. Primary use: Generative molecular design. Platform/source: Hugging Face.

Generative molecular designtrained modelchemistry modelgenerative designdiffusion
Model · Generative molecular design

NVIDIA MegaMolBART

Transformer model for molecular representation learning, interpolation, and SMILES-based molecule generation. Model kind: model family. Primary use: Generative molecular design. Platform/source: NVIDIA BioNeMo.

Generative molecular designtrained modelchemistry modelgenerative designSMILES
Model · Generative molecular design

NVIDIA MolMIM

Mutual-information machine model for molecular generation, optimization, and latent-space molecule editing. Model kind: model family. Primary use: Generative molecular design. Platform/source: NVIDIA BioNeMo.

Generative molecular designtrained modelchemistry modelgenerative designlatent optimization
Model · Protein-ligand docking and binding

P2Rank binding-site model

Machine-learning model for ligand binding-site prediction from protein structure. Model kind: model system. Primary use: Protein-ligand docking and binding. Platform/source: GitHub.

Protein-ligand docking and bindingtrained modelchemistry modelbinding siteprotein structure
Model · Generative molecular design

PS-VAE

Molecule generation by principal subgraph mining and assembling. Model kind: model family. Primary use: Generative molecular design. Platform/source: GitHub.

Generative molecular designtrained modelchemistry modelsubgraph generationVAE
Model · Generative molecular design

PhoreGen

Pharmacophore-oriented 3D molecular generation model. Model kind: model family. Primary use: Generative molecular design. Platform/source: GitHub.

Generative molecular designtrained modelchemistry modelpharmacophore3D generation
Model · Generative molecular design

QizhiPei/biot5-plus-base-mol-instructions-molecule

BioT5+ molecule-instruction model for molecule-text tasks and molecular reasoning. Model kind: checkpoint. Primary use: Generative molecular design. Platform/source: Hugging Face.

Generative molecular designtrained modelchemistry modelBioT5molecule instructions
Model · Generative molecular design

REINVENT4 prior and agent models

AI molecular design system for de novo design, scaffold hopping, R-group replacement, linker design, and molecule optimization. Model kind: model system. Primary use: Generative molecular design. Platform/source: GitHub.

Generative molecular designtrained modelchemistry modelgenerative designde novo design
Model · Retrosynthesis and reaction prediction

RXNMapper atom-mapping model

Unsupervised attention-guided atom-mapping model for chemical reactions. Model kind: model system. Primary use: Retrosynthesis and reaction prediction. Platform/source: GitHub.

Retrosynthesis and reaction predictiontrained modelchemistry modelatom mappingreaction grammar
Model · Generative molecular design

TargetDiff

3D equivariant diffusion model for target-aware molecule generation and affinity prediction. Model kind: model family. Primary use: Generative molecular design. Platform/source: GitHub.

Generative molecular designtrained modelchemistry modeldiffusiontarget aware
Model · Protein-ligand docking and binding

TorchDrug pretrained drug-target models

Drug-discovery model framework with pretrained workflows for molecular and protein tasks. Model kind: model system. Primary use: Protein-ligand docking and binding. Platform/source: GitHub.

Protein-ligand docking and bindingtrained modelchemistry modeldrug discoveryprotein ligand
Model · Generative molecular design

Torsional Diffusion

Torsional diffusion model for molecular conformer generation. Model kind: model family. Primary use: Generative molecular design. Platform/source: GitHub.

Generative molecular designtrained modelchemistry modeldiffusionconformer generation
Model · Protein-ligand docking and binding

UniSite binding-site detection

Protein-ligand binding-site detection checkpoint. Model kind: checkpoint. Primary use: Protein-ligand docking and binding. Platform/source: Hugging Face.

Protein-ligand docking and bindingtrained modelchemistry modelbinding siteprotein ligand
Model · Generative molecular design

alimotahharynia/DrugGen

DrugGen model fine-tuned from DrugGPT for approved drug-target style generation. Model kind: checkpoint. Primary use: Generative molecular design. Platform/source: Hugging Face.

Generative molecular designtrained modelchemistry modelgenerative designdrug target
Model · Generative molecular design

blazerye/DrugAssist-7B

Large language model for molecule generation and molecule optimization in drug discovery. Model kind: checkpoint. Primary use: Generative molecular design. Platform/source: Hugging Face.

Generative molecular designtrained modelchemistry modelgenerative designLLM
Model · Generative molecular design

chandar-lab/NovoMolGen_157M_SMILES_BPE

NovoMolGen 157M SMILES BPE checkpoint for drug-like molecular generation. Model kind: checkpoint. Primary use: Generative molecular design. Platform/source: Hugging Face.

Generative molecular designtrained modelchemistry modelgenerative designSMILES
Model · Generative molecular design

chandar-lab/NovoMolGen_300M_SMILES_AtomWise

NovoMolGen 300M atomwise tokenizer variant for molecule generation. Model kind: checkpoint. Primary use: Generative molecular design. Platform/source: Hugging Face.

Generative molecular designtrained modelchemistry modelgenerative designSMILES
Model · Generative molecular design

chandar-lab/NovoMolGen_300M_SMILES_BPE

NovoMolGen 300M SMILES BPE model for molecular generation trained on ZINC-22. Model kind: checkpoint. Primary use: Generative molecular design. Platform/source: Hugging Face.

Generative molecular designtrained modelchemistry modelgenerative designSMILES
Model · Generative molecular design

chandar-lab/NovoMolGen_32M_SMILES_BPE

Compact NovoMolGen 32M BPE model for SMILES molecular generation. Model kind: checkpoint. Primary use: Generative molecular design. Platform/source: Hugging Face.

Generative molecular designtrained modelchemistry modelgenerative designSMILES
Model · Property prediction and ADMET

dptech/Uni-Mol-Models

Uni-Mol pretrained models for 3D molecular representation learning. Model kind: checkpoint. Primary use: Property prediction and ADMET. Platform/source: Hugging Face.

Property prediction and ADMETtrained modelchemistry modelUni-Mol3D molecule
Model · Property prediction and ADMET

dptech/Uni-Mol2

Uni-Mol2 model family for 3D molecular representation learning. Model kind: checkpoint. Primary use: Property prediction and ADMET. Platform/source: Hugging Face.

Property prediction and ADMETtrained modelchemistry modelUni-Mol23D molecule
Model · Generative molecular design

hogru/MolReactGen-GuacaMol-Molecules

GPT-2 style model for GuacaMol molecule generation tasks. Model kind: checkpoint. Primary use: Generative molecular design. Platform/source: Hugging Face.

Generative molecular designtrained modelchemistry modelGuacaMolSMILES
Model · Generative molecular design

ibm-research/GP-MoLFormer-Uniq

Generative pre-trained MoLFormer model for molecule generation from molecular language representations. Model kind: checkpoint. Primary use: Generative molecular design. Platform/source: Hugging Face.

Generative molecular designtrained modelchemistry modelgenerative designMoLFormer
Model · Property prediction and ADMET

ibm-research/MoLFormer-XL-both-10pct

MoLFormer XL molecular foundation model for feature extraction and property prediction. Model kind: checkpoint. Primary use: Property prediction and ADMET. Platform/source: Hugging Face.

Property prediction and ADMETtrained modelchemistry modelMoLFormermolecular foundation model
Model · Property prediction and ADMET

ibm-research/biomed.sm.mv-te-84m

IBM small-molecule multiview 84M base model used for MoleculeNet finetunes. Model kind: model family. Primary use: Property prediction and ADMET. Platform/source: Hugging Face.

Property prediction and ADMETtrained modelchemistry modelsmall moleculeIBM
Model · Structure recognition and utility models

knowledgator/SMILES2IUPAC-canonical-base

SMILES-to-IUPAC translation model for canonical chemical naming. Model kind: checkpoint. Primary use: Structure recognition and utility models. Platform/source: Hugging Face.

Structure recognition and utility modelstrained modelchemistry modelSMILESIUPAC
Model · Generative molecular design

laituan245/molt5-large-smiles2caption

MolT5 large model translating SMILES to molecule captions. Model kind: checkpoint. Primary use: Generative molecular design. Platform/source: Hugging Face.

Generative molecular designtrained modelchemistry modelMolT5SMILES caption
Model · Generative molecular design

language-plus-molecules/mCLM_1k-3b

Molecule-conditioned language model for molecule generation and multimodal molecular tasks. Model kind: checkpoint. Primary use: Generative molecular design. Platform/source: Hugging Face.

Generative molecular designtrained modelchemistry modelmolecular language modelmultimodal
Model · Generative molecular design

language-plus-molecules/molt5-large-caption2smiles-LPM24

MolT5 large caption-to-SMILES model for text-guided molecule generation. Model kind: checkpoint. Primary use: Generative molecular design. Platform/source: Hugging Face.

Generative molecular designtrained modelchemistry modelcaption to SMILEStext guided generation
Model · Generative molecular design

liyuesen/druggpt

DrugGPT checkpoint for GPT-style de novo molecule generation. Model kind: checkpoint. Primary use: Generative molecular design. Platform/source: Hugging Face.

Generative molecular designtrained modelchemistry modelgenerative designGPT
Model · Protein-ligand docking and binding

nvidia/NV-Proteina-Complexa-Ligand-Target-160M-v1

Protein-ligand target model from NVIDIA for complex-level binding tasks. Model kind: checkpoint. Primary use: Protein-ligand docking and binding. Platform/source: Hugging Face.

Protein-ligand docking and bindingtrained modelchemistry modelprotein ligandNVIDIA
Model · Retrosynthesis and reaction prediction

sagawa/ReactionT5v1-forward

ReactionT5 v1 model for forward product prediction. Model kind: checkpoint. Primary use: Retrosynthesis and reaction prediction. Platform/source: Hugging Face.

Retrosynthesis and reaction predictiontrained modelchemistry modelforward reactionReactionT5
Model · Retrosynthesis and reaction prediction

sagawa/ReactionT5v1-retrosynthesis

ReactionT5 v1 model for retrosynthesis prediction. Model kind: checkpoint. Primary use: Retrosynthesis and reaction prediction. Platform/source: Hugging Face.

Retrosynthesis and reaction predictiontrained modelchemistry modelretrosynthesisReactionT5
Model · Retrosynthesis and reaction prediction

sagawa/ReactionT5v1-yield

ReactionT5 v1 reaction yield model. Model kind: checkpoint. Primary use: Retrosynthesis and reaction prediction. Platform/source: Hugging Face.

Retrosynthesis and reaction predictiontrained modelchemistry modelyield predictionReactionT5
Model · Retrosynthesis and reaction prediction

sagawa/ReactionT5v2-forward

ReactionT5 v2 checkpoint for forward reaction product prediction. Model kind: checkpoint. Primary use: Retrosynthesis and reaction prediction. Platform/source: Hugging Face.

Retrosynthesis and reaction predictiontrained modelchemistry modelforward reactionproduct prediction
Model · Retrosynthesis and reaction prediction

sagawa/ReactionT5v2-forward-USPTO_MIT

ReactionT5 v2 forward model trained on USPTO-MIT style reaction data. Model kind: checkpoint. Primary use: Retrosynthesis and reaction prediction. Platform/source: Hugging Face.

Retrosynthesis and reaction predictiontrained modelchemistry modelforward reactionUSPTO-MIT
Model · Retrosynthesis and reaction prediction

sagawa/ReactionT5v2-retrosynthesis

ReactionT5 v2 checkpoint for retrosynthesis using reaction SMILES and ORD-derived training. Model kind: checkpoint. Primary use: Retrosynthesis and reaction prediction. Platform/source: Hugging Face.

Retrosynthesis and reaction predictiontrained modelchemistry modelretrosynthesisReactionT5
Model · Retrosynthesis and reaction prediction

sagawa/ReactionT5v2-retrosynthesis-USPTO_50k

ReactionT5 v2 retrosynthesis model trained on USPTO-50K. Model kind: checkpoint. Primary use: Retrosynthesis and reaction prediction. Platform/source: Hugging Face.

Retrosynthesis and reaction predictiontrained modelchemistry modelretrosynthesisUSPTO-50K
Model · Retrosynthesis and reaction prediction

sagawa/ReactionT5v2-yield

ReactionT5 v2 model for reaction yield prediction. Model kind: checkpoint. Primary use: Retrosynthesis and reaction prediction. Platform/source: Hugging Face.

Retrosynthesis and reaction predictiontrained modelchemistry modelyield predictionReactionT5
Model · Property prediction and ADMET

seyonec/ChemBERTa-zinc-base-v1

Widely used ChemBERTa ZINC pretraining checkpoint for molecular representations. Model kind: checkpoint. Primary use: Property prediction and ADMET. Platform/source: Hugging Face.

Property prediction and ADMETtrained modelchemistry modelChemBERTaZINC
Model · Structure recognition and utility models

seyonec/PubChem10M_SMILES_BPE_450k

SMILES BPE pretraining checkpoint built from PubChem10M. Model kind: checkpoint. Primary use: Structure recognition and utility models. Platform/source: Hugging Face.

Structure recognition and utility modelstrained modelchemistry modelSMILESPubChem
Model · Structure recognition and utility models

unikei/bert-base-smiles

BERT model pretrained on SMILES for molecular representation learning. Model kind: checkpoint. Primary use: Structure recognition and utility models. Platform/source: Hugging Face.

Structure recognition and utility modelstrained modelchemistry modelSMILESBERT
Model · Generative molecular design

zjunlp/MolGen

Domain-agnostic molecular generation with chemical feedback. Model kind: model family. Primary use: Generative molecular design. Platform/source: GitHub.

Generative molecular designtrained modelchemistry modelmolecular generationchemical feedback
Model · Property prediction

ALIGNN: Atomistic Line Graph Neural Network

The Atomistic Line Graph Neural Network (ALIGNN) is a deep learning model for predicting material properties. It leverages graph neural networks to represent atomic structures and their relationships, enabling accurate property predictions.

materials sciencedeep learninggraph neural networksmaterials predictionatomistic modeling
Model

OrbMol-v2: Learnable Per-Atom Electrostatics Model

OrbMol-v2 is an extension of the OrbMol architecture, trained on the Open Molecules 2025 (OMol25) and OPoly26 datasets. It incorporates learnable per-atom electrostatics, including latent charge prediction and a long-range Coulomb energy m...

materials sciencemachine learningmolecular modelingelectrostaticsdeep learning
Model

A Large Encoder-Decoder Family of Foundation Models For Chemical Language

This paper introduces a large encoder-decoder chemical foundation model pre-trained on 91 million SMILES samples (4 billion molecular tokens) from PubChem. The model supports tasks like quantum property prediction and offers variants with...

chemical language modelsfoundation modelsencoder-decodercheminformaticsSMILES
Model

NequIP: Machine Learning Interatomic Potentials

NequIP is an open-source framework for developing accurate, fast, and scalable machine learning interatomic potentials, featuring various pre-trained models and extensions.

machine learninginteratomic potentialsNequIPmaterials scienceGNN
Model · Generative materials design

MatterGen: Generative Model for Inorganic Materials Design

Official implementation of MatterGen, a generative model for designing inorganic materials across the periodic table. It can be fine-tuned to meet specific property constraints.

materials designgenerative modelinorganic materialsAImachine learning