{"success":true,"data":{"author":"Axelrod and Gomez-Bombarelli (MIT); mirror by raulsofia","bucketAccessType":"public","bucketName":"raulsofia/geom_drugs","bucketPrefix":"","category":"Chemistry & Materials","cloudProvider":"huggingface","createdAt":"2026-11-16T02:38:31.237Z","description":"The drug-like subset of GEOM (Geometric Ensemble Of Molecules) from the Gomez-Bombarelli group at MIT: conformer ensembles of about 300,000 drug-like molecules with associated experimental data, generated with CREST metadynamics at the semi-empirical GFN2-xTB level, with conformer energies and Boltzmann statistical weights. This mirror packages the conformers and SMILES as SDF and text files in two versions: allconfs/ with every available conformer and 5confs/ with at most five per molecule. Each holds an unsplit *_confs.sdf, *_counts.csv (conformers per SMILES) and *_smiles.smi, plus an 80/20 train/test split made on unique SMILES so that no molecule appears in both; about 73 GB in total (measured 2026-11-15). Read with RDKit SDMolSupplier. GEOM-Drugs is the standard benchmark for 3D molecular generation and conformer generation models (EDM, GeoDiff, Torsional Diffusion, MiDi). The original GEOM release on Harvard Dataverse (doi:10.7910/DVN/JNGTDF) is CC0-1.0; cite Axelrod and Gomez-Bombarelli, Scientific Data 9, 185 (2022).","displayTitle":"GEOM-Drugs: Energy-Annotated Conformer Ensembles for Drug-Like Molecules (SDF)","id":"c27527a7-a115-515f-b584-d03ec5755eac","license":"CC0-1.0","licenseEvidenceUrl":"https://dataverse.harvard.edu/dataset.xhtml?persistentId=doi:10.7910/DVN/JNGTDF","objectCountEstimate":20,"probeStatus":"ok","referenceUrl":"https://doi.org/10.1038/s41597-022-01288-4","region":"us","schemaCard":{"accessHints":[{"language":"python-datasets","snippet":"from datasets import load_dataset\nds = load_dataset(\"raulsofia/geom_drugs\")"},{"language":"python-hub","snippet":"from huggingface_hub import snapshot_download\nsnapshot_download(repo_id=\"raulsofia/geom_drugs\", repo_type=\"dataset\")"}],"caveats":["HuggingFace metadata: author=raulsofia, license:cc-by-4.0, size_categories:100K<n<1M, format:csv, downloads=1184"],"format":"mixed","generatedAt":"2026-11-16T02:38:39.582Z","layout":{"example":"5confs/5confs_counts.csv","pattern":"{repoId}/{filename}"},"mountPath":"/mnt/archive","sampleHeaders":[{"bytes":"SMILES,Count\r\n[H][C](/[C]([Br])=[N]/[N]([H])[c]1[n][c]([N]2[C]([H])([H])[C]([H])([H])[O][C]([H])([H])[C]2([H])[H])[n][c]([N]2[C]([H])([H])[C]([H])([H])[O][C]([H])([H])[C]2([H])[H])[n]1)=[C](\\[H])[c]1[c]([H])[c]([H])[c]([H])[c]([H])[c]1[H],5\r\n[H][C](=[N]\\[N]1[C]([H])([H])[C]([H])([H])[N]([C]([H])([H])[c]2[c]([H])[c]([H])[c]([H])[c]([H])[c]2[H])[C]([H])([H])[C]1([H])[H])/[C]([Br])=[C](/[H])[c]1[c]([H])[c]([H])[c]([H])[c]([H])[c]1[H],5\r\n[H][C](=[N]\\[N]([H])[c]1[n][c]([N]2[C]([H])([H])[C]([H])([H])[O][C]([H])([","encoding":"utf8","key":"5confs/5confs_counts.csv","truncatedAt":512}],"sampleObjects":["5confs/5confs_confs.sdf","5confs/5confs_counts.csv","5confs/5confs_smiles.smi","5confs/test/test_confs.sdf","5confs/test/test_counts.csv"],"sidecars":[{"bytes":"---\nlicense: cc-by-4.0\ntags:\n- chemistry\n- conformer-generation\n- molecules\npretty_name: GEOM-Drugs (Mirror)\n---\n\n# GEOM: Molecular Conformations (Drugs Subset)\n\n**Note:** *This is a mirrored and specifically preprocessed version of the GEOM dataset (Drugs subset), originally created by Simon Axelrod and Rafael Gómez-Bombarelli. All credit for the original conformational sampling and DFT calculations goes to the original authors. This repository exists to guarantee availability and exact reproducibility for downstream machine learning projects.*\n\n## Dataset Description\nThe Geometric Ensemble Of Molecules (GEOM) dataset contains millions of molecular conformations annotated by energy and statistical weight. This specific repository hosts the \"Drugs\" subset, which focuses on realistically sized drug-like molecules with experimental data, and contains only the conformers and smiles associated.\n\n* **Original Paper:** [GEOM, energy-annotated molecular conformations for property prediction and molecular generation](https://www.nature.com/articles/s41597-022-01288-4)\n* **Original GitHub:** [learningmatter-mit/geom](https://github.com/learningmatter-mit/geom)\n\n---\n\n## Data Structure & Organization\n\nTo maximize flexibility and reproducibility, this repository is organized into specific dataset versions and pre-computed splits. \n\n### Available Versions\n* **`allconfs/`**: The full, unredacted version containing all available conformers.\n* **`5confs/`**: A redacted, lightweight version limited to a maximum of 5 conformers per molecule.\n\n### Directory Layout\nInside each version folder, you will find the **unsplit** dataset files located at the root, alongside pre-computed `train/` and `test/` subdirectories:\n\n```text\ngeom_drugs/\n├── 5confs/\n│   ├── train/\n│   │   ├── train_confs.sdf\n│   │   ├── train_counts.csv\n│   │   └── train_smiles.smi\n│   ├── test/\n│   │   ├── test_confs.sdf\n│   │   ├── test_counts.csv\n│   │   └── test_smiles.smi\n│   ├── 5confs_confs.sdf   <-- Unsplit dataset\n│   ├── 5confs_counts.csv   <-- Unsplit counts\n│   └── 5confs_smiles.smi   <-- Unsplit SMILES\n└── allconfs/\n    └── ... (Matches the structure above)\n```\n\n### File Types\nEach split (and the unsplit root) contains three complementary files representing the exact same pool of molecules:\n1. **`*_confs.sdf`**: The main Structure-Data File containing the 3D molecular conformers.\n2. **`*_counts.csv`**: A summary file mapping each unique SMILES string to the number of conformers it has in the corresponding `.sdf` file.\n3. **`*_smiles.smi`**: A strict 1:1 text file containing the SMILES string for every single entry in the `.sdf` file (useful for fast, sequential parsing or redundancy checks).\n\n---\n\n## Train / Test Splitting Strategy\n\nThe `train/` and `test/` subfolders contain an 80/20 split of the data. \n\n**Note on Data Leakage:** The splits are calculated strictly based on **unique SMILES keys**, *not* by individual conformer counts. This guarantees that all conformers belonging to a specific molecule are assigned exclusively to either the training set or the testing set. There are absolutely **zero shared molecules** between the train and test splits, ensuring robust validation of model generalizability. \n\nIf you prefer to use alternative splitting strategies (e.g., scaffold splits or specific property cutoffs), you can ignore the subdirectories and build custom splits using the unsplit files located at the root of each version folder.\n\n---\n\n## Citation Information\nIf you use this dataset, you must cite the original creators:\n\n```bibtex\n@article{axelrod2022geom,\n  title={GEOM, energy-annotated molecular conformations for property prediction and molecular generation},\n  author={Axelrod, Simon and G{\\'o}mez-Bombarelli, Rafael},\n  journal={Scientific Data},\n  volume={9},\n  number={1},\n  pages={185},\n  year={2022},\n  publisher={Nature Publishing Group UK London}\n}\n```","encoding":"utf8","key":"README.md","name":"README.md","size":4014,"truncated":false}],"sourceObjectKey":"5confs/5confs_counts.csv","starterCell":"import pandas as pd\ndf = pd.read_csv(\"hf://datasets/raulsofia/geom_drugs/5confs/5confs_counts.csv\")\ndf.head()","starterSamplePath":"5confs/5confs_counts.csv","totalObjects":20},"summary":"The drug-like subset of GEOM (Geometric Ensemble Of Molecules) from the Gomez-Bombarelli group at MIT: conformer ensembles of about 300,000 drug-like molecules with associated…","title":"GEOM-Drugs: Energy-Annotated Conformer Ensembles for Drug-Like Molecules (SDF)","updatedAt":"2026-11-16T17:04:50.982Z","visibility":"public","websiteUrl":"https://github.com/learningmatter-mit/geom"}}