Wiley Interdisciplinary Reviews: Computational Molecular Science最新文献

Machine learning (ML) techniques offer a novel and exciting approach in the drug discovery field. One might even argue that their current expansion may push traditional MM modeling techniques to a secondary role in modeling methods. In this review article, we advocate that a combination of both techniques could be the most efficient implementation in the coming years. Focusing on drug-target affinity predictions, we first review pure ML approaches. Then, we introduced recent developments in mixing ML and MM methods in a single combined manner. Finally, we show the detailed implementation of a real industrial prospective study where nanomolar hits, on a kinase target, were obtained by combination of state of the art Monte Carlo MM simulations (PELE) with a ML ranking function.

This article is categorized under:

Electronic structure calculations of many-electron systems should in principle treat relativistic, correlation, and quantum electrodynamics (QED) effects simultaneously to a high precision, so as to match experimental measurements as close as possible. While both relativistic and QED effects can readily be built into the many-electron Hamiltonian, electron correlation is more difficult to describe due to the exponential growth of the number of parameters in the wave function. Compared with the spin-free case, spin–orbit interaction results in the loss of spin symmetry and concomitant complex algebra, thereby rendering the treatment of electron correlation even more difficult. Possible solutions to these issues are highlighted here.

This article is categorized under:

Brownian dynamics (BD) is a computational method to simulate molecular diffusion processes. Although the BD method has been developed over several decades and is well established, new methodological developments are improving its accuracy, widening its scope, and increasing its application. In biological applications, BD is used to investigate the diffusive behavior of molecules subject to forces due to intermolecular interactions or interactions with material surfaces. BD can be used to compute rate constants for diffusional association, generate structures of encounter complexes for molecular binding partners, and examine the transport properties of geometrically complex molecules. Often, a series of simulations is performed, for example, for different protein mutants or environmental conditions, so that the effects of the changes on diffusional properties can be estimated. While biomolecules are commonly described at atomic resolution and internal molecular motions are typically neglected, coarse-graining and the treatment of conformational flexibility are increasingly employed. Software packages for BD simulations of biomolecules are growing in capabilities, with several new packages providing novel features that expand the range of questions that can be addressed. These advances, when used in concert with experiment or other simulation methods, such as molecular dynamics, open new opportunities for application to biochemical and biological systems. Here, we review some of the latest developments in the theory, methods, software, and applications of BD simulations to study biomolecular diffusional association processes and provide a perspective on their future use and application to outstanding challenges in biology, bioengineering, and biomedicine.

This article is categorized under:

Quantum mechanical (QM) calculations are critical in quantitatively understanding the relationship between the structure and physicochemical properties of various chemical systems. However, the sharply increasing computational cost with the system size has severely hindered applying direct QM calculations on large-sized systems. Hence, linear-scaling and/or fragmentation QM methods have been proposed to overcome this difficulty. In this review, we focus on the recent development and applications of the electrostatically embedded generalized molecular fractionation with the conjugate caps (EE-GMFCC) method in probing various properties of complex large molecules and condensed-phase systems. The EE-GMFCC method is now capable of describing the localized excited states of biomolecules and molecular crystals with a chromophore. The EE-GMF method is also combined with anharmonic vibrational calculations for accurate simulation of the infrared spectrum of the magic number H⁺(H₂O)₂₁ cluster at the coupled cluster level. With an adaptive fragmentation scheme, the EE-GMF-based ab initio molecular dynamics is able to directly simulate chemical reactions occurred in atmospheric molecular clusters. Furthermore, by combining the EE-GMF(CC) method and deep machine learning techniques, neural network potentials can be efficiently constructed for accurate simulations of complex systems with the accuracy of high-level wave function methods. The EE-GMF(CC) method is expected to become a practical tool for quantitative description of complex large molecules and condensed-phase systems with high-level ab initio theories or ab initio quality potentials.

This article is categorized under:

SERENITY [J Comput Chem. 2018;39:788–798] is an open-source quantum chemistry software that provides an extensive development platform focused on quantum-mechanical multilevel and embedding approaches. In this study, we give an overview over the developments done in Serenity since its original publication in 2018. This includes efficient electronic-structure methods for ground states such as multilevel domain-based local pair natural orbital coupled cluster and Møller–Plesset perturbation theory as well as the multistate frozen-density embedding quasi-diabatization method. For the description of excited states, SERENITY features various subsystem-based methods such as embedding variants of coupled time-dependent density-functional theory, approximate second-order coupled cluster theory and the second-order algebraic diagrammatic construction technique as well as GW/Bethe–Salpeter equation approaches. SERENITY's modular structure allows combining these methods with density-functional theory (DFT)-based embedding through various practical realizations and variants of subsystem DFT including frozen-density embedding, potential-reconstruction techniques and projection-based embedding.

This article is categorized under:

The ONIOM method and many extensions to it provide capabilities to treat challenging multiscale problems in catalysis and material science. Our open-source program Spicy is a flexible toolkit for ONIOM and fragment methods. Spicy includes a generalization of multicenter-ONIOM, a higher-order multipole embedding scheme, and fragment methods as useful extensions of our own n-layered integrated molecular orbital and molecular mechanics (ONIOM), which allow applying ONIOM and high accuracy calculations to a wider range of systems. A calculation on the metallo-protein hemoglobin demonstrates the versatility of the implementation.

This article is categorized under:

The power of computer simulations, including machine-learning, has become an inseparable part of scientific analysis of biological data. This has significantly impacted the field of cryogenic electron microscopy (cryo-EM), which has grown dramatically since the “resolution-revolution.” Many maps are now solved at 3–4 Å or better resolution, although a significant proportion of maps deposited in the Electron Microscopy Data Bank are still at lower resolution, where the positions of atoms cannot be determined unambiguously. Additionally, cryo-EM maps are often characterized by a varying local resolution, partly due to conformational heterogeneity of the imaged molecule. To address such problems, many computational methods have been developed for cryo-EM map reconstruction and atomistic model building. Here, we review the development in algorithms and tools for building models in cryo-EM maps at different resolutions. We describe methods for model building, including rigid and flexible fitting of known models, model validation, small-molecule fitting, and model visualization. We provide examples of how these methods have been used to elucidate the structure and function of dynamic macromolecular machines.

This article is categorized under:

The cover image is based on the Focus Article New phase space formulations and quantum dynamics approaches by Xin He et al., https://doi.org/10.1002/wcms.1619.

Machine learning techniques have been widely applied in many fields of chemistry, physics, biology, and materials science. One of the most fruitful applications is machine learning of the complicated multidimensional function of potential energy or related electronic properties from discrete quantum chemical data. In particular, substantial efforts have been dedicated to developing various atomistic neural network (AtNN) representations, which refer to a family of methods expressing the targeted physical quantity as a sum of atomic components represented by atomic NNs. This class of approaches not only fully preserves the physical symmetry of the system but also scales linearly with respect to the size of a system, enabling accurate and efficient chemical dynamics and spectroscopic simulations in complicated systems and even a number of variably sized systems across the phases. In this review, we discuss different strategies in developing highly efficient and representable AtNN potentials, and in generalizing these scalar AtNN models to learn vectorial and tensorial quantities with the correct rotational equivariance. We also review active learning algorithms to generate practical AtNN models and present selected examples of AtNN applications in gas-surface systems to demonstrate their capabilities of accurately representing both molecular systems and condensed phase systems. We conclude this review by pointing out remaining challenges for the further development of more reliable, transferable, and scalable AtNN representations in more application scenarios.

This article is categorized under:

A fundamental and challenging task of computational protein studies is to design proteins of desired structures and functions on demand. Data-driven approaches to protein design have been gaining tremendous momentum, with recent developments concentrated on protein sequence representation and generation by using deep learning language models, structure-based sequence design or inverse protein folding, and the de novo generation of new protein backbones. Currently, design methods have been assessed mainly by several useful computational metrics. However, these metrics are still highly insufficient for predicting the performance of design methods in wet experiments. Nevertheless, some methods have been verified experimentally, which showed that proteins of novel sequences and structures can be designed with data-driven models learned from natural proteins. Despite the progress, an important current limitation is the lack of accurate data-driven approaches to model or design protein dynamics.

This article is categorized under: