nablaDFT: Large-Scale Conformational Energy and Hamiltonian Prediction benchmark and dataset

Khrabrov, Kuzma; Shenbin, Ilya; Ryabov, Alexander; Tsypin, Artem; Telepov, Alexander; Alekseev, Anton; Grishin, Alexander; Страшнов, П. В.; Zhilyaev, Petr; Nikolenko, Sergey I.; Kadurin, Artur

doi:10.1039/d2cp03966d

Cited by 8 publications

(6 citation statements)

References 67 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…So-called semiempirical methods 10 , 11 , such as the Neglect of Diatomic Orbital Overlap (NDDO 12 ) and modern related methods 13 , tight-binding DFT (xTB 14 ), or ‘quantum force fields’ 15 , can greatly improve the scaling of traditional quantum chemistry methods through simplification or approximation of the underlying physics, without restrictions on the domain of applicability imposed by purely data-driven methods. Nonetheless, there is great interest in supporting this vision by taking a purely data-driven approach and developing quantum machine learning (QML) models based on accurate physical methods such as density functional theory (DFT) combined with large and diverse collections of data (e.g., QM9 16 , QMugs 17 , nablaDFT 18 and QM7-X 19 ). Another approach is to devise transfer learning datasets and algorithms that can extract useful patterns from less accurate, but cheaper and more scalable simulations that ultimately benefit predictions at a higher fidelity level 8 , 17 , 20 , 21 .…”

Section: Introductionmentioning

confidence: 99%

Modelling local and general quantum mechanical properties with attention-based pooling

Buterez,

Janet,

Kiddle

et al. 2023

Commun Chem

View full text Add to dashboard Cite

Atom-centred neural networks represent the state-of-the-art for approximating the quantum chemical properties of molecules, such as internal energies. While the design of machine learning architectures that respect chemical principles has continued to advance, the final atom pooling operation that is necessary to convert from atomic to molecular representations in most models remains relatively undeveloped. The most common choices, sum and average pooling, compute molecular representations that are naturally a good fit for many physical properties, while satisfying properties such as permutation invariance which are desirable from a geometric deep learning perspective. However, there are growing concerns that such simplistic functions might have limited representational power, while also being suboptimal for physical properties that are highly localised or intensive. Based on recent advances in graph representation learning, we investigate the use of a learnable pooling function that leverages an attention mechanism to model interactions between atom representations. The proposed pooling operation is a drop-in replacement requiring no changes to any of the other architectural components. Using SchNet and DimeNet++ as starting models, we demonstrate consistent uplifts in performance compared to sum and mean pooling and a recent physics-aware pooling operation designed specifically for orbital energies, on several datasets, properties, and levels of theory, with up to 85% improvements depending on the specific task.

show abstract

Section: Introductionmentioning

confidence: 99%

Modelling local and general quantum mechanical properties with attention-based pooling

Buterez,

Janet,

Kiddle

et al. 2023

Commun Chem

View full text Add to dashboard Cite

show abstract

“…This chemical space is incomplete, as many drug molecules contain phosphorus, sulfur, and halogen atoms, and some contain metal ions. − The ANI-2x model was extended to include S, F, and Cl, but the full data set, including the important reference energies and forces at the ωB97X/6-31G* level, to our knowledge, has not yet become publicly available. Currently, there are a number of recent data sets that include compounds that contain phosphorus, sulfur, and halogens at various levels of theory − as well as metal ions . Among them, only the SPICE data set includes forces at the ωB97M-D3BJ/def2-TZVPPD level and currently includes over 420K phosphorus, 520K sulfur, 750K halogen, and 8K metal-containing structures.…”

Section: Resultsmentioning

confidence: 99%

QDπ: A Quantum Deep Potential Interaction Model for Drug Discovery

Zeng

Tao

Giese

et al. 2023

J. Chem. Theory Comput.

View full text Add to dashboard Cite

We report QDπ-v1.0 for modeling the internal energy of drug molecules containing H, C, N, and O atoms. The QDπ model is in the form of a quantum mechanical/machine learning potential correction (QM/Δ-MLP) that uses a fast third-order self-consistent density-functional tight-binding (DFTB3/3OB) model that is corrected to a quantitatively high-level of accuracy through a deep-learning potential (DeepPot-SE). The model has the advantage that it is able to properly treat electrostatic interactions and handle changes in charge/protonation states. The model is trained against reference data computed at the ωB97X/6-31G* level (as in the ANI-1x data set) and compared to several other approximate semiempirical and machine learning potentials (ANI-1x, ANI-2x, DFTB3, MNDO/d, AM1, PM6, GFN1-xTB, and GFN2-xTB). The QDπ model is demonstrated to be accurate for a wide range of intra- and intermolecular interactions (despite its intended use as an internal energy model) and has shown to perform exceptionally well for relative protonation/deprotonation energies and tautomers. An example application to model reactions involved in RNA strand cleavage catalyzed by protein and nucleic acid enzymes illustrates QDπ has average errors less than 0.5 kcal/mol, whereas the other models compared have errors over an order of magnitude greater. Taken together, this makes QDπ highly attractive as a potential force field model for drug discovery.

show abstract

“…Several quantum chemistry (QC) data sets have been made publicly available, − but some are less suitable for pretraining drug-like molecules. For example, QM9, 54 Alchemy, nablaDFT and QM1B have limited ranges of molecular weights and do not sufficiently cover the chemical space of drug-like molecules. SPICE is more leaned toward application for machine learned potentials.…”

Section: Related Workmentioning

confidence: 99%

Quantum-Informed Molecular Representation Learning Enhancing ADMET Property Prediction

Kim,

Chang,

et al. 2024

J. Chem. Inf. Model.

View full text Add to dashboard Cite

nablaDFT: Large-Scale Conformational Energy and Hamiltonian Prediction benchmark and dataset

Abstract: In this work we present nablaDFT, the new dataset and benchmark for the Density Functional Theory Hamiltonian and energy prediction. We provide data for over 1 million different molecules and over 5 million conformations and baseline models for both tasks.

Cited by 8 publications

References 67 publications

Modelling local and general quantum mechanical properties with attention-based pooling

Modelling local and general quantum mechanical properties with attention-based pooling

QDπ: A Quantum Deep Potential Interaction Model for Drug Discovery

Quantum-Informed Molecular Representation Learning Enhancing ADMET Property Prediction

Contact Info

Product

Resources

About