Note Browse datasets by property, application, and modality. Find loading examples, source citations, licenses, and usage notes in each dataset card.

Note PCQM4Mv2 PubChemQC HOMO-LUMO gap benchmark. Large-scale prediction of quantum-calculated HOMO–LUMO gaps from molecular structures. Includes the official train, validation, test-development, and test-challenge partitions, with a separate archive of training geometries.

Note QM9 equilibrium quantum molecular properties. Quantum-chemical properties of 133,885 small organic molecules, with molecular structures, calculated electronic and thermodynamic targets, and corresponding 3D geometries.

Note ChEMBL 37 full bioactivity database. ChEMBL 37 molecular structures and relational bioactivity records, linked to assays, targets, measurement types, units, and original documents.

Note CheMixHub miscible-solvent. CheMixHub data for simulated density, mixing, and vaporization properties of solvent mixtures, with component molecular structures, mixture composition, source conditions, and property labels. Each label describes the complete mixture.

Note MassSpecGym 1.5 MS/MS molecule-discovery benchmark. MassSpecGym 1.5 pairs tandem mass spectra with molecular identities for molecular discovery and retrieval. Includes spectrum metadata, benchmark folds, and links to the matching mass- and formula-based candidate pools.

Note OpenADMET ExpansionRx ADMET challenge. Experimental ADMET challenge data from ExpansionRx and OpenADMET, including molecular structures, endpoint labels, and available assay and quality-control information.

Note ASAP–Polaris–OpenADMET property challenge. Experimental property-prediction challenge data from ASAP, Polaris, and OpenADMET. Molecular tables include source assay context; crystallographic structures are available as a separate archive.

Note OpenADMET CYP challenge. OpenADMET cytochrome P450 challenge data with molecular structures, experimental endpoint labels, and source training/test partitions.

Note OpenADMET PXR challenge. OpenADMET pregnane X receptor challenge data with molecular structures, experimental endpoint labels, and source training/test partitions.

Note OpenADMET pKa modeling data. Experimental acid-dissociation data for molecular pKa modeling, with source molecular structures and available measurement context.

Note PI1M v2. Approximately one million generated polymer repeat units, with heuristic synthetic accessibility scores in v2; no experimental property labels.

Note RadonPy PI1070. Original RadonPy public table for amorphous polymers with simulation and calculation metadata.

Note PolyOmics. PolyOmics combines polymer property tables with molecular-dynamics snapshot archives. Records cover multiple polymer architectures and simulation conditions; snapshot filenames use the property table’s UUID identifiers.

Note PolyID. NREL (now National Laboratory of the Rockies) public constructed-polymer training input with experimental Tg and Tm labels, stereochemistry and monomer metadata.

Note SMiPoly virtual library. Rule-generated virtual polymer repeat units and 1,083 curated monomer inputs released by the original authors.

Note PolyMetriX curated glass transition temperatures. Curated experimental Tg dataset with source, reliability, dispersion, and polymer metadata; canonical original Zenodo v1.

Note PolyBench26 / OMersBench curated MD source. PolyBench26 curated 82,990-record MD source with Cp, Cv, Rg, density and refractive index; architecture metadata is preserved.

Note VIPEA polymer ensemble electronic properties. Aldeghi and Coley polymer ensemble electronic-property data, preserving copolymer architecture and monomer components.

Note Open Macromolecular Genome quantum property labels. Original active-learning and test tables containing quantum-calculated capped-monomer properties for synthetically accessible OMG repeat units.

Note Tox21 nuclear receptor and stress response assays. Tox21 nuclear receptor and stress response assays. DeepChem/MoleculeNet release.

Note ToxCast high-throughput mechanistic toxicity assays. ToxCast high-throughput mechanistic toxicity assays. DeepChem/MoleculeNet release.

Note ChEMBL2K multi-target activity panel. About 2,000 molecules with activity labels on 41 protein targets, curated from ChEMBL for InfoAlign.

Note Broad6K cell-based assay panel. About 6,000 molecules with Broad Institute cell-based assay readouts, curated for InfoAlign.

Note TDC ADMET multitask table (ADMET-AI recipe). The 41 TDC ADMET benchmark endpoints merged into one table, one row per molecule, following ADMET-AI's merge recipe; released with torch-molecule.

Note ZINC250k. About 250,000 drug-like molecules sampled from ZINC with computed logP, QED, and SA score, released with the chemical variational autoencoder.

Note Polymer gas permeability (CH4, CO2, H2, N2, O2). Gas permeability of 786 polymers for five gases, released with torch-molecule.

Note Harvard Organic Photovoltaic dataset (HOPV15). Complete DeepChem experimental HOPV table, including small molecules and polymers. Preserves device architecture, acceptor complement, source DOI, material construction and the supplied molecular representation. The original HOPV publication also provides computed properties; those are not part of this processed CSV snapshot.

Note QMugs full molecular property summary. Calculated molecular and conformer properties for more than 665,000 ChEMBL compounds and about two million conformers. Includes the property summary and the corresponding SDF conformer archive.

Note ClinTox clinical toxicity and FDA approval. ClinTox clinical toxicity and FDA approval. DeepChem/MoleculeNet release.

Note FreeSolv experimental hydration free energy. FreeSolv experimental hydration free energy. DeepChem/MoleculeNet release.

Note SIDER drug side effects by organ class. SIDER drug side effects by organ class. DeepChem/MoleculeNet release.

Note ESOL / Delaney aqueous solubility. ESOL / Delaney aqueous solubility. DeepChem/MoleculeNet release.

Note BBBP blood-brain barrier permeability. BBBP blood-brain barrier permeability. DeepChem/MoleculeNet release.

Note AstraZeneca lipophilicity / logD. AstraZeneca lipophilicity / logD. DeepChem/MoleculeNet release.

Note NCI HIV antiviral screen. NCI HIV antiviral screen. DeepChem/MoleculeNet release.

Note BACE-1 inhibitor potency and binary activity. BACE-1 inhibitor potency and binary activity. DeepChem/MoleculeNet release.

Note Thermodynamic solubility in pH 7.4 buffer. Thermodynamic solubility in pH 7.4 buffer. DeepChem/MoleculeNet release.

Note QM7 molecular atomization energy (DeepChem SMILES view). Atomization-energy regression for small organic molecules with up to seven heavy atoms. This SMILES benchmark contains 6,834 molecules; the full 7,165-record matrix reference is listed separately.

Note Hepatocyte clearance. Hepatocyte clearance. DeepChem/MoleculeNet release.

Note MUV maximum unbiased validation virtual screening. MUV maximum unbiased validation virtual screening. DeepChem/MoleculeNet release.

Note QM8 excited-state electronic spectra. Electronic excitation energies and oscillator strengths for small organic molecules, calculated with several electronic-structure methods. Corrected PBE0 labels come from the original geometry-linked property table; method-specific columns remain separate.

Note NCI-60 cancer cell-line growth inhibition. NCI-60 cancer cell-line growth inhibition. DeepChem/MoleculeNet release.

Note Caco-2 permeability (TDC). Molecular structures and source labels for Caco-2 permeability, distributed through Therapeutics Data Commons.

Note PAMPA permeability (TDC). Molecular structures and source labels for PAMPA permeability, distributed through Therapeutics Data Commons.

Note Human intestinal absorption (TDC). Molecular structures and source labels for human intestinal absorption, distributed through Therapeutics Data Commons.

Note P-glycoprotein inhibition (TDC). Molecular structures and source labels for P-glycoprotein inhibition, distributed through Therapeutics Data Commons.

Note Oral bioavailability (TDC). Molecular structures and source labels for oral bioavailability, distributed through Therapeutics Data Commons.

Note Steady-state volume of distribution (TDC). Molecular structures and source labels for steady-state volume of distribution, distributed through Therapeutics Data Commons.

Note Elimination half-life (TDC). Molecular structures and source labels for elimination half-life, distributed through Therapeutics Data Commons.

Note Hepatocyte clearance (TDC). Molecular structures and source labels for hepatocyte clearance, distributed through Therapeutics Data Commons.

Note Plasma protein binding (TDC). Molecular structures and source labels for plasma protein binding, distributed through Therapeutics Data Commons.

Note Microsome clearance (TDC). Molecular structures and source labels for microsome clearance, distributed through Therapeutics Data Commons.

Note Human liver microsomal stability (TDC). Molecular structures and source labels for human liver microsomal stability, distributed through Therapeutics Data Commons.

Note Rat liver microsomal stability (TDC). Molecular structures and source labels for rat liver microsomal stability, distributed through Therapeutics Data Commons.

Note HERG channel blockade (TDC). Molecular structures and source labels for hERG channel blockade, distributed through Therapeutics Data Commons.

Note Drug-induced liver injury (TDC). Molecular structures and source labels for drug-induced liver injury, distributed through Therapeutics Data Commons.

Note Skin sensitization (TDC). Molecular structures and source labels for skin sensitization, distributed through Therapeutics Data Commons.

Note HERG blockade at 10 uM threshold (TDC). Molecular structures and source labels for hERG blockade at 10 uM threshold, distributed through Therapeutics Data Commons.

Note Carcinogenicity (TDC). Molecular structures and source labels for carcinogenicity, distributed through Therapeutics Data Commons.

Note Ames mutagenicity (TDC). Molecular structures and source labels for Ames mutagenicity, distributed through Therapeutics Data Commons.

Note Acute rat oral LD50 (TDC). Molecular structures and source labels for acute rat oral LD50, distributed through Therapeutics Data Commons.

Note SARS-CoV-2 3CL protease screening (TDC). Molecular structures and source labels for SARS-CoV-2 3CL protease screening, distributed through Therapeutics Data Commons.

Note SARS-CoV-2 in-vitro antiviral activity (TDC). Molecular structures and source labels for SARS-CoV-2 in-vitro antiviral activity, distributed through Therapeutics Data Commons.

Note HERG central screen (TDC). Molecular structures and source labels for hERG central screen, distributed through Therapeutics Data Commons.

Note QM7b multilevel electronic properties. Multiple calculated electronic properties for the QM7b small-molecule collection, provided in the original numerical representation.

Note PubChem BioAssay 128-assay molecular screening benchmark. PubChem BioAssay 128-assay molecular screening benchmark. DeepChem/MoleculeNet release.

Note Polaris-curated Fang/Biogen prospective ADME. Prospective ADME benchmark covering six in-vitro drug-metabolism and pharmacokinetic endpoints collected over 20 months. This release contains 3,521 source rows with assay context and quality information.

Note CARA assay-aware compound activity benchmark v1. Original CARA v1 archive, assay-aware activity data and benchmark split definitions.

Note Octant CYP inhibition and reactivity. Octant CYP inhibition and reactivity measurements with compound identifiers, molecular structures, and assay readouts. Reactivity labels describe depletion derived from control and treatment peak areas.

Note OpenPoly 2025. Literature-curated polymer property table, distributed with the OpenPoly benchmark; preserve missing values and source property names.

Note ToPoRg-1342. Single-chain squared radii of gyration for 1,342 topologically diverse coarse-grained polymers. Carbon SMILES encode graph topology only.

Note Polymer–solvent Crystal16 solubility. Processed cooling-cycle turbidity measurements with separate original polymer and solvent SMILES mapping tables.

Note HTP-MD / PolyGen public training subset. Toyota Research Institute released subset of high-throughput polymer-electrolyte MD, used in PolyGen training.

Note PolyUniverse virtual polymer library. Virtual polymer candidates with neural-network property predictions. This release covers the PBI, polyamide, and polyester class tables; it supports computational screening rather than experimental property validation.

Note AqSolDB curated aqueous solubility. Aqueous-solubility regression data compiled from nine earlier sources. Includes molecular structures, solubility labels, and source reliability and reference fields.

Note BDE-db2 and external bond-energy tests. Bond-specific energy prediction data with molecular structures, bond indices, and radical fragments. The main training labels are calculated; an experimental external test is identified separately.

Note SPICE 2.0.1. SPICE 2.0.1 provides atomistic configurations of drug-like molecules, peptides, ion pairs, and solvated systems, with quantum-calculated energies and gradients at ωB97M-D3BJ/def2-TZVPPD. The original HDF5 representation preserves atom ordering and geometry-linked labels.

Note NMRGym molecular structure elucidation. NMRGym spectrum benchmark with paired proton/carbon NMR data and scaffold-based partitions. Native Python pickle files preserve the released spectrum records and their source representation.

Note MolecularIQ graph-level molecular reasoning. MolecularIQ evaluates molecular graph reasoning through task-specific inputs and symbolic answers. Includes the released task configurations and training pool; generation tasks may not have a fixed input SMILES.

Note MolPuzzle molecular structure-elucidation benchmark. MolPuzzle combines molecular questions and answers with spectrum interpretation and molecule-construction tasks. Available spectrum images remain linked to their source molecule and modality identifiers.

Note ChemPile chemistry pretraining corpus. ChemPile is a multimodal chemistry corpus containing text, molecular captions, images, code, papers, and structured chemical records. This repository provides eight original components, totaling about 193 GB, for chemistry language and multimodal modeling.

Note ChemX chemistry information-extraction datasets. ChemX benchmarks chemical information extraction from literature across nanomaterials and small molecules. Records retain extraction context and source article identifiers; the original article PDFs are not included.

Note Open Polymers 2026 (OPoly26). OPoly26 polymer-cluster configurations for learning atomistic energies and forces. This repository contains the original train/validation archive from release 260302, including geometries, charge, and spin information in native atomistic databases.

Note FGBench functional-group molecular reasoning

Note NeurIPS Open Polymer Challenge 2025

Note B3DB blood-brain barrier permeability (TDC)

Note Butkiewicz PubChem HTS benchmark (TDC)

Note CheMixHub mixture property benchmark

Note CYP450 substrates (Carbon-Mangels, TDC)

Note CYP450 inhibition (Veith 2009, TDC)

Note LLNL CHNO-1 10K energetic molecules

Note MARCEL conformer-ensemble benchmark

Note MoleculeACE activity-cliff benchmark

Note MolLangBench (ICLR 2026)

Note OPERA physicochemical, environmental-fate and toxicity data

Note polyVERSE polymer property database

Note SimPoly polymer benchmark

Note ToxiMol structure-level detoxification

Note WelQrate curated virtual screening