GeoMol: Torsional Geometric Generation of Molecular 3D Conformer Ensembles

Authors: Octavian Ganea, Lagnajit Pattanaik, Connor Coley, Regina Barzilay, Klavs Jensen, William Green, Tommi Jaakkola

NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Empirically, we conduct experiments on two benchmarks: GEOM-QM9 (smaller molecules relevant to gas-phase chemistry) and GEOM-DRUGS (drug-like molecules) [Axelrod and Gomez Bombarelli, 2020a]. Our method often outperforms previous ML and two popular open-source or commercial methods in different metrics.
Researcher Affiliation Academia Department of Chemical Engineering, MIT, Cambridge, MA 02139 Computer Science and Artificial Intelligence Laboratory, MIT, Cambridge, MA 02139
Pseudocode No No pseudocode or algorithm blocks are explicitly presented in the paper.
Open Source Code Yes Code is available at https://github.com/Pattanaik L/Geo Mol.
Open Datasets Yes We use two popular datasets: GEOM-QM9 [Ramakrishnan et al., 2014] and GEOM-DRUGS [Axelrod and Gomez-Bombarelli, 2020a].
Dataset Splits Yes We split them randomly based on molecules into train/validation/test (80%/10%/10%).
Hardware Specification No No specific hardware specifications (e.g., GPU/CPU models, memory) are mentioned for the experimental setup.
Software Dependencies No No specific software dependencies with version numbers are mentioned.
Experiment Setup Yes At train time, our model uses a standard deviation (std) s (see eq. (1)) of 5 for both GEOM-QM9 and GEOM-DRUGS. For GEOM-QM9, this corresponds to no RMSD cutoff (i.e., OMEGA generates all possible conformers), and for GEOM-DRUGS, this corresponds to a cutoff of 0.7Å. We discuss hyper-parameters and additional training details in appendix H.