GEOM-QM9 & GEOM-DRUGS preprocessed conformer pickles (Align Your Structures, ICLR 2026)

Fuente: Zenodo
Salvato in:
Dettagli Bibliografici
Autori principali: Iyengar, Aniketh, Han, Jiaqi, Sun, Pengwei, Jiang, Mingjian, Xie, Jianwen, Ermon, Stefano
Natura: Recurso digital
Lingua:inglese
Pubblicazione: Zenodo 2026
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866901639009402880
author Iyengar, Aniketh
Han, Jiaqi
Sun, Pengwei
Jiang, Mingjian
Xie, Jianwen
Ermon, Stefano
author_facet Iyengar, Aniketh
Han, Jiaqi
Sun, Pengwei
Jiang, Mingjian
Xie, Jianwen
Ermon, Stefano
contents <p>Preprocessed GEOM-QM9 and GEOM-DRUGS conformer pickles used for conformer-level diffusion pretraining (Stage 1) in <em>Align Your Structures: Generating Trajectories with Structure Pretraining for Molecular Dynamics</em> (ICLR 2026).</p><p><strong>Contents (6 pickles, ~3 GB uncompressed):</strong></p><ul><li><code>GEOM-QM9_Train.pkl</code> — QM9 train split</li><li><code>GEOM-QM9_Val.pkl</code> — QM9 validation split</li><li><code>GEOM-QM9_Test_Actual_compat.pkl</code> — QM9 test split (ConfGF 200-molecule official benchmark, repackaged)</li><li><code>GEOM-DRUGS_Train.pkl</code> — DRUGS train split</li><li><code>GEOM-DRUGS_Val.pkl</code> — DRUGS validation split</li><li><code>GEOM-DRUGS_Test_Actual_compat.pkl</code> — DRUGS test split (ConfGF 200-molecule official benchmark, repackaged)</li></ul><p><strong>Format:</strong> each pickle is a Python <code>list</code> of <code>dict</code>s with keys <code>atom_type</code>, <code>boltzmannweight</code>, <code>edge_index</code>, <code>edge_type</code>, <code>idx</code>, <code>nx</code>, <code>pos</code>, <code>rdmol</code>, <code>smiles</code>, <code>totalenergy</code>. Same fields as GeoDiff's PyTorch-Geometric <code>Data</code> format, repackaged as dicts to decouple loading from PyG version drift.</p><p><strong>Provenance:</strong> train / val derived from the GeoDiff preprocessed GEOM archive (<a href="https://github.com/MinkaiXu/GeoDiff">MinkaiXu/GeoDiff</a>), itself built on top of the ConfGF preprocessing pipeline (<a href="https://github.com/DeepGraphLearning/ConfGF">DeepGraphLearning/ConfGF</a>). Test split is ConfGF's official 200-molecule benchmark. Upstream raw GEOM: Axelrod & Gómez-Bombarelli, <em>Scientific Data</em> 2022, doi:10.7910/DVN/JNGTDF.</p><p><strong>Extraction:</strong></p><pre>tar xf align-your-structures-conformer-pkls-v1.tar.gz -C ${MD_DATA_ROOT}/</pre><p>The archive expands into <code>processed_input_data/GEOM-{QM9,DRUGS}/</code>, matching the paths referenced in the <code>configs_official/</code> YAMLs.</p><p><strong>Reference code:</strong> <a href="https://github.com/ani11452/Align_Your_Structures">https://github.com/ani11452/Align_Your_Structures</a></p><p>If you use this data please cite the paper above and the upstream GEOM dataset.</p>
format Recurso digital
id zenodo_https___doi_org_10_5281_zenodo_19711753
institution Zenodo
language eng
publishDate 2026
publisher Zenodo
record_format zenodo
spellingShingle GEOM-QM9 & GEOM-DRUGS preprocessed conformer pickles (Align Your Structures, ICLR 2026)
Iyengar, Aniketh
Han, Jiaqi
Sun, Pengwei
Jiang, Mingjian
Xie, Jianwen
Ermon, Stefano
molecular conformers
GEOM-QM9
GEOM-DRUGS
diffusion models
generative models
ICLR 2026
<p>Preprocessed GEOM-QM9 and GEOM-DRUGS conformer pickles used for conformer-level diffusion pretraining (Stage 1) in <em>Align Your Structures: Generating Trajectories with Structure Pretraining for Molecular Dynamics</em> (ICLR 2026).</p><p><strong>Contents (6 pickles, ~3 GB uncompressed):</strong></p><ul><li><code>GEOM-QM9_Train.pkl</code> — QM9 train split</li><li><code>GEOM-QM9_Val.pkl</code> — QM9 validation split</li><li><code>GEOM-QM9_Test_Actual_compat.pkl</code> — QM9 test split (ConfGF 200-molecule official benchmark, repackaged)</li><li><code>GEOM-DRUGS_Train.pkl</code> — DRUGS train split</li><li><code>GEOM-DRUGS_Val.pkl</code> — DRUGS validation split</li><li><code>GEOM-DRUGS_Test_Actual_compat.pkl</code> — DRUGS test split (ConfGF 200-molecule official benchmark, repackaged)</li></ul><p><strong>Format:</strong> each pickle is a Python <code>list</code> of <code>dict</code>s with keys <code>atom_type</code>, <code>boltzmannweight</code>, <code>edge_index</code>, <code>edge_type</code>, <code>idx</code>, <code>nx</code>, <code>pos</code>, <code>rdmol</code>, <code>smiles</code>, <code>totalenergy</code>. Same fields as GeoDiff's PyTorch-Geometric <code>Data</code> format, repackaged as dicts to decouple loading from PyG version drift.</p><p><strong>Provenance:</strong> train / val derived from the GeoDiff preprocessed GEOM archive (<a href="https://github.com/MinkaiXu/GeoDiff">MinkaiXu/GeoDiff</a>), itself built on top of the ConfGF preprocessing pipeline (<a href="https://github.com/DeepGraphLearning/ConfGF">DeepGraphLearning/ConfGF</a>). Test split is ConfGF's official 200-molecule benchmark. Upstream raw GEOM: Axelrod & Gómez-Bombarelli, <em>Scientific Data</em> 2022, doi:10.7910/DVN/JNGTDF.</p><p><strong>Extraction:</strong></p><pre>tar xf align-your-structures-conformer-pkls-v1.tar.gz -C ${MD_DATA_ROOT}/</pre><p>The archive expands into <code>processed_input_data/GEOM-{QM9,DRUGS}/</code>, matching the paths referenced in the <code>configs_official/</code> YAMLs.</p><p><strong>Reference code:</strong> <a href="https://github.com/ani11452/Align_Your_Structures">https://github.com/ani11452/Align_Your_Structures</a></p><p>If you use this data please cite the paper above and the upstream GEOM dataset.</p>
title GEOM-QM9 & GEOM-DRUGS preprocessed conformer pickles (Align Your Structures, ICLR 2026)
topic molecular conformers
GEOM-QM9
GEOM-DRUGS
diffusion models
generative models
ICLR 2026
url https://doi.org/10.5281/zenodo.19711753