SDDF Energy Dataset

This conformational energy dataset, developed as part of the Smart Distributed Data Factory (SDDF) project, contains over 2.75 million molecular conformations based on drug-like molecules sourced from the ENAMINE database. Energies were calculated using DFT with the ωB97x density functional and the 6–31G(d) basis set. The conformations were generated from SMILES using RDKit, MMFF94 optimization, and molecular dynamics (MD) simulations, providing a diverse set of molecular structures and energy states.

• RDKit Conformations: 1,123,693
• RDKit + MMFF94 Optimized: 1,151,936
• MD-Generated: 483,279

This dataset serves as a benchmark for energy prediction models, with training (638,617 examples), validation (134,732 examples), and test subsets (24,890 examples) created using a strict scaffold-based split to ensure no overlap and less than 70% similarity between the training and test sets.

Dataset contents:

• data.tar.gz: contains the conformations in Structured Data File format, grouped into separate folders based on the molecule ID. Each conformation's label is provided within its SDF file as a property named "energy".
• INDEX.smi: specifies the molecule IDs and their corresponding SMILES.
• SOURCES.csv: specifies the conformation generation method for each conformation.
• SDDF_train.tsv, SDDF_validation.tsv, and SDDF_test.tsv specify the molecule IDs and conformations for each subset of the benchmark.

A detailed description is provided in the accompanying paper.