QM datasets#
The physicsml package uses the molflux.datasets module which provides efficient and convenient handling
of datasets and direct access to publicly available ones (such as ANI1ccx). It is built on top of HuggingFace datasets.
As part of the molflux package, you can access multiple quantum mechanical datasets directly. The dataset files are
downloaded from source, processed on the fly, and the dataset is constructed and cached for later use. For more information
about molflux datasets, please see the docs. All of the datasets below require a kwarg specifying the
backend to process the molecules, either 'rdkit' or 'openeye'
from molflux.datasets import load_dataset
dataset = load_dataset("dataset_name", "backend_name")
ani1x#
The ANI-1x and ANI-1ccx ML-based general-purpose datasets for organic molecules were developed through active learning; an automated data diversification process. The ANI-1x data set contains multiple quantum mechanical properties from 5M density functional theory calculations, while the ANI-1ccx data set contains 500k data points obtained with an accurate CCSD(T)/CBS extrapolation. Approximately 14 million CPU core-hours were expended to generate this data. Multiple QM calculated properties for the chemical elements C, H, N, and O are provided: energies, atomic forces, multipole moments, atomic charges, etc.
ani2x#
The new model, dubbed ANI-2x, is trained to three additional chemical elements: S, F, and Cl. Additionally, ANI-2x underwent torsional refinement training to better predict molecular torsion profiles. These new features open a wide range of new applications within organic chemistry and drug development. These seven elements (H, C, N, O, F, Cl, and S) make up ~90% of drug-like molecules.
gdb9#
Computational de novo design of new drugs and materials requires rigorous and unbiased exploration of chemical compound space. However, large uncharted territories persist due to its size scaling combinatorially with molecular size. We report computed geometric, energetic, electronic, and thermodynamic properties for 134k stable small organic molecules made up of CHONF. These molecules correspond to the subset of all 133,885 species with up to nine heavy atoms (CONF) out of the GDB-17 chemical universe of 166 billion organic molecules. We report geometries minimal in energy, corresponding harmonic frequencies, dipole moments, polarizabilities, along with energies, enthalpies, and free energies of atomization. All properties were calculated at the B3LYP/6-31G(2df,p) level of quantum chemistry. Furthermore, for the predominant stoichiometry, C7H10O2, there are 6,095 constitutional isomers among the 134k molecules. We report energies, enthalpies, and free energies of atomization at the more accurate G4MP2 level of theory for all of them. As such, this data set provides quantum chemical properties for a relevant, consistent, and comprehensive chemical space of small organic molecules. This database may serve the benchmarking of existing methods, development of new methods, such as hybrid quantum mechanics/machine learning, and systematic identification of structure-property relationships.
Note
Description from source.
spice#
SPICE dataset, a new quantum chemistry dataset for training potentials relevant to simulating drug-like small molecules interacting with proteins. It contains over 1.1 million conformations for a diverse set of small molecules, dimers, dipeptides, and solvated amino acids. It includes 15 elements, charged and uncharged molecules, and a wide range of covalent and non-covalent interactions. It provides both forces and energies calculated at the ωB97M-D3(BJ)/def2-TZVPPD level of theory.
Note
Description from source.
The molecules here are point clouds. They do not have any bonds. The formal charges are added from the SMILES.
pcqm4m_v2#
PCQM4Mv2 is a quantum chemistry dataset originally curated under the PubChemQC project. Based on the PubChemQC, we define a meaningful ML task of predicting DFT-calculated HOMO-LUMO energy gap of molecules given their 2D molecular graphs. The HOMO-LUMO gap is one of the most practically-relevant quantum chemical properties of molecules since it is related to reactivity, photoexcitation, and charge transport. Moreover, predicting the quantum chemical property only from 2D molecular graphs without their 3D equilibrium structures is also practically favorable. This is because obtaining 3D equilibrium structures requires DFT-based geometry optimization, which is expensive on its own.
We provide molecules as the SMILES strings, from which 2D molecule graphs (nodes are atoms and edges are chemical bonds). We further provide the equilibrium 3D graph structure for training molecules.
Note
Description from source.