Ab Initio Machine Learning in Chemical Compound Space

Chemical Reviews, Journal Year: 2021, Volume and Issue: 121(16), P. 10001 - 10036

Published: Aug. 13, 2021

Chemical compound space (CCS), the set of all theoretically conceivable combinations chemical elements and (meta-)stable geometries that make up matter, is colossal. The first-principles based virtual sampling this space, for example, in search novel molecules or materials which exhibit desirable properties, therefore prohibitive but smallest subsets simplest properties. We review studies aimed at tackling challenge using modern machine learning techniques on (i) synthetic data, typically generated quantum mechanics methods, (ii) model architectures inspired by mechanics. Such Quantum Machine Learning (QML) approaches combine numerical efficiency statistical surrogate models with an ab initio view matter. They rigorously reflect underlying physics order to reach universality transferability across CCS. While state-of-the-art approximations problems impose severe computational bottlenecks, recent QML developments indicate possibility substantial acceleration without sacrificing predictive power

Language: Английский

---

Inverse design of 3d molecular structures with conditional generative neural networks

Nature Communications, Journal Year: 2022, Volume and Issue: 13(1)

Published: Feb. 21, 2022

The rational design of molecules with desired properties is a long-standing challenge in chemistry. Generative neural networks have emerged as powerful approach to sample novel from learned distribution. Here, we propose conditional generative network for 3d molecular structures specified chemical and structural properties. This agnostic bonding enables targeted sampling distributions, even domains where reference calculations are sparse. We demonstrate the utility our method inverse by generating motifs or composition, discovering particularly stable molecules, jointly targeting multiple electronic beyond training regime.

Language: Английский

---

Representations of molecules and materials for interpolation of quantum-mechanical simulations via machine learning

npj Computational Materials, Journal Year: 2022, Volume and Issue: 8(1)

Published: March 16, 2022

Computational study of molecules and materials from first principles is a cornerstone physics, chemistry, science, but limited by the cost accurate precise simulations. In settings involving many simulations, machine learning can reduce these costs, often orders magnitude, interpolating between reference This requires representations that describe any molecule or material support interpolation. We comprehensively review discuss current relations them, using unified mathematical framework based on many-body functions, group averaging, tensor products. For selected state-of-the-art representations, we compare energy predictions for organic molecules, binary alloys, Al-Ga-In sesquioxides in numerical experiments controlled data distribution, regression method, hyper-parameter optimization.

Language: Английский

---

Accurate global machine learning force fields for molecules with hundreds of atoms

Science Advances, Journal Year: 2023, Volume and Issue: 9(2)

Published: Jan. 11, 2023

Global machine learning force fields, with the capacity to capture collective interactions in molecular systems, now scale up a few dozen atoms due considerable growth of model complexity system size. For larger molecules, locality assumptions are introduced, consequence that nonlocal not described. Here, we develop an exact iterative approach train global symmetric gradient domain (sGDML) fields (FFs) for several hundred atoms, without resorting any potentially uncontrolled approximations. All atomic degrees freedom remain correlated sGDML FF, allowing accurate description complex molecules and materials present phenomena far-reaching characteristic correlation lengths. We assess accuracy efficiency on newly developed MD22 benchmark dataset containing from 42 370 atoms. The robustness our is demonstrated nanosecond path-integral dynamics simulations supramolecular complexes dataset.

Language: Английский

---

Quantum machine learning for chemistry and physics

Chemical Society Reviews, Journal Year: 2022, Volume and Issue: 51(15), P. 6475 - 6573

Published: Jan. 1, 2022

Machine learning (ML) has emerged into formidable force for identifying hidden but pertinent patterns within a given data set with the objective of subsequent generation automated predictive behavior. In recent years, it is safe to conclude that ML and its close cousin deep (DL) have ushered unprecedented developments in all areas physical sciences especially chemistry. Not only classical variants , even those trainable on near-term quantum hardwares been developed promising outcomes. Such algorithms revolutionzed material design performance photo-voltaics, electronic structure calculations ground excited states correlated matter, computation force-fields potential energy surfaces informing chemical reaction dynamics, reactivity inspired rational strategies drug designing classification phases matter accurate identification emergent criticality. this review we shall explicate subset such topics delineate contributions made by both computing enhanced machine over past few years. We not present brief overview well-known techniques also highlight their using statistical insight. The foster exposition aforesaid empower promote cross-pollination among future-research chemistry which can benefit from turn potentially accelerate growth algorithms.

Language: Английский

---