Journal of Chemical Theory and Computation最新文献

This paper presents a grid-based approach to model molecular association processes as an alternative to sampling-based Markov models. Our method discretizes the six-dimensional space of relative translation and orientation into grid cells. By discretizing the Fokker-Planck operator governing the system dynamics via the square-root approximation, we derive analytical expressions for the transition rate constants between grid cells. These expressions depend on geometric properties of the grid, such as the cell surface area and volume, which we provide. In addition, one needs only the molecular energy at the grid cell center, circumventing the need for extensive MD simulations and reducing the number of energy evaluations to the number of grid cells. The resulting rate matrix is closely related to the Markov state model transition matrix, offering insights into metastable states and association kinetics. We validate the accuracy of the model in identifying metastable states and binding mechanisms, though improvements are necessary to address limitations like ignoring bulk transitions and anisotropic rotational diffusion. The flexibility of this grid-based method makes it applicable to a variety of molecular systems and energy functions, including those derived from quantum mechanical calculations. The software package MolGri, which implements this approach, offers a systematic and computationally efficient tool for studying molecular association processes.

Local hybrid functionals (LHs) use a real-space position-dependent admixture of exact exchange (EXX), governed by a local mixing function (LMF). The systematic construction of LMFs has been hampered over the years by a lack of exact physical constraints on their valence behavior. Here, we exploit a data-driven approach and train a new type of "n-LMF" as a relatively shallow neural network. The input features are of meta-GGA character, while the W4-17 atomization-energy and BH76 reaction-barrier test sets have been used for training. Simply replacing the widely used "t-LMF" of the LH20t functional by the n-LMF provides the LH24n-B95 functional. Augmented by DFT-D4 dispersion corrections, LH24n-B95-D4 remarkably improves the WTMAD-2 value for the large GMTKN55 test suite of general main-group thermochemistry, kinetics, and noncovalent interactions (NCIs) from 4.55 to 3.49 kcal/mol. As we found the limited flexibility of the B95c correlation functional to disfavor much further improvement on NCIs, we proceeded to replace it by an optimized B97c-type power-series expansion. This gives the LH24n functional. LH24n-D4 gives a WTMAD-2 value of 3.10 kcal/mol, the so far lowest value of a rung 4 functional in self-consistent calculations. The new functionals perform moderately well for organometallic transition-metal energetics while leaving room for further data-driven improvements in that area. Compared to complete neural-network functionals like DM21, the present more tailored approach to train just the LMF in a flexible but well-defined human-designed LH functional retains the possibility of graphical LMF analyses to gain deeper understanding. We find that both the present n-LMF and the recent x-LMF suppress the so-called gauge problem of local hybrids without adding a calibration function as required for other LMFs. LMF plots show that this can be traced back to large LMF values in the small-density region between the interacting atoms in NCIs for n- and x-LMFs and low values for the t-LMF. We also find that the trained n-LMF has relatively large values in covalent bonds without deteriorating binding energies. The current approach enables fast and efficient routine self-consistent calculations using n-LMFs in Turbomole.

We have developed a method that uses energy landscapes of unbound and bound ligands to compute reorganization free energies for end-point binding free-energy calculations. The method is applied to our previous simulations of fentanyl derivatives bound to the μ opioid receptor in different orientations. Whereas the mean interaction energy provides an ambiguous ranking of binding poses, interaction entropy and ligand reorganization strongly penalize geometric decoys such that native poses observed in CryoEM structures are best ranked. The binding pose of fentanyl is driven by the interaction entropy. Binding of (3R,4S)-lofentanil is favored over that of (3S,4R)-lofentanil, largely because binding the latter requires the ligand to reorganize to a conformation with high free energy. The same phenomenon is predicted to favor the binding orientation of carfentanil. Our method can be applied to other end-point binding free-energy calculations for a relatively low cost of sampling the unbound ligand. Source code is included in the Supporting Information.

Accurate structural feature characterization of cyclic peptides (CPs), especially those with less than 10 residues and cis-peptide bonds, is challenging but important for the rational design of bioactive peptides. In this study, we performed high-temperature molecular dynamics (high-T MD) simulations on 250 CPs with random sequences and applied the point-adaptive k-nearest neighbors (PAk) method to estimate the free energies of millions of sampled conformations. Using this data set, we trained a SchNet-based deep learning model, termed CPconf_score, to predict the conformational free energies of CPs. We tested CPconf_score to identify near-native conformations from MD-sampled conformations of 50 CPs from the Cambridge Structural Database. Our method achieved accurate predictions for 41 out of 50 CPs with a backbone RMSD of less than 1.0 Å compared to crystal structures. In comparison, other advanced CP structure prediction tools, such as HighFold and Rosetta, successfully predicted 12 and 19 CPs, respectively.

The cost of encoding a system Hamiltonian in a digital quantum computer as a linear combination of unitaries (LCU) grows with the 1-norm of the LCU expansion. The Block Invariant Symmetry Shift (BLISS) technique reduces this 1-norm by modifying the Hamiltonian action on only the undesired electron-number subspaces. Previously, BLISS required a computationally expensive nonlinear optimization that was not guaranteed to find the global minimum. Here, we introduce various reformulations of this optimization as a linear programming problem, which guarantees optimality and significantly reduces the computational cost. We apply BLISS to industrially relevant homogeneous catalysts in active spaces of up to 76 orbitals, finding substantial reductions in both the spectral range of the modified Hamiltonian and the 1-norms of Pauli and fermionic LCUs. Our linear programming techniques for obtaining the BLISS operator enable more efficient Hamiltonian simulation and, by reducing the Hamiltonian's spectral range, offer opportunities for improved LCU groupings to further reduce the 1-norm.

While machine learning (ML) models have been able to achieve unprecedented accuracies across various prediction tasks in quantum chemistry, it is now apparent that accuracy on a test set alone is not a guarantee for robust chemical modeling such as stable molecular dynamics (MD). To go beyond accuracy, we use explainable artificial intelligence (XAI) techniques to develop a general analysis framework for atomic interactions and apply it to the SchNet and PaiNN neural network models. We compare these interactions with a set of fundamental chemical principles to understand how well the models have learned the underlying physicochemical concepts from the data. We focus on the strength of the interactions for different atomic species, how predictions for intensive and extensive quantum molecular properties are made, and analyze the decay and many-body nature of the interactions with interatomic distance. Models that deviate too far from known physical principles produce unstable MD trajectories, even when they have very high energy and force prediction accuracy. We also suggest further improvements to the ML architectures to better account for the polynomial decay of atomic interactions.

Traditional machine learning methods face significant challenges in predicting the properties of highly symmetric molecules. In this study, we developed a machine learning model based on graph neural networks (GNNs) to accurately and swiftly predict the thermodynamic and photochemical properties of fullerenols, such as C₆₀(OH)_n (n = 1 to 30). First, we established a global method for generating fullerenol isomers through isomer fingerprinting, which can generate all possible isomers or produce diverse structural types on demand. Significantly, by incorporating interpretable descriptors such as atomic labels, bond lengths, and bond angles from highly symmetric isomers, our multilayer GNN model achieved over 90% accuracy in predicting the thermodynamic stability of fullerenols. The model also performed excellently in predicting electronic properties, including the highest occupied molecular orbital (HOMO), lowest unoccupied molecular orbital (LUMO), and the energy gap. Overall, this work demonstrates a new strategy using interpretable descriptors for accurately predicting the properties of highly symmetric structures, offering theoretical chemists a valuable tool for studying these materials.

Embedding techniques allow the efficient description of correlations within localized fragments of large molecular systems while accounting for their environment at a lower level of theory. We introduce FragPT2: a novel embedding framework that addresses multiple interacting active fragments. Fragments are assigned separate active spaces, constructed by localizing canonical molecular orbitals. Each fragment is then solved with a multireference method, self-consistently embedded in the mean field from other fragments. Finally, interfragment correlations are reintroduced through multireference perturbation theory. Our framework provides an exhaustive classification of interfragment interaction terms, offering a tool to analyze the relative importance of various processes such as dispersion, charge transfer, and spin exchange. We benchmark FragPT2 on challenging test systems, including N₂ dimers, multiple aromatic dimers, and butadiene. We demonstrate that our method can be successful even for fragments defined by cutting through a covalent bond.

The accuracy and efficiency of a coarse-grained (CG) force field are pivotal for high-precision molecular simulations of large systems with complex molecules. We present an automated mapping and optimization framework for molecular simulation (AMOFMS), which is designed to streamline and improve the force field optimization process. It features a neural-network-based mapping function, DSGPM-TP (deep supervised graph partitioning model with type prediction). This model can accurately and efficiently convert atomistic structures to CG mappings, reducing the need for manual intervention. By integrating bottom-up and top-down methodologies, AMOFMS allows users to freely combine these approaches or use them independently as optimization targets. Moreover, users can select and combine different optimizers to meet their specific mission. With its parallel optimizer, AMOFMS significantly accelerates the optimization process, reducing the time required to achieve optimal results. Successful applications of AMOFMS include parameter optimizations for systems such as POPC and PEO, demonstrating its robustness and effectiveness. Overall, AMOFMS provides a general and flexible framework for the automated development of high-precision CG force fields.

Our recently developed approach based on the local coupled-cluster with single, double, and perturbative triple excitation [LCCSD(T)] model gives very efficient means to compute the ideal-gas enthalpies of formation. The expanded uncertainty (95% confidence) of the method is about 3 kJ·mol^-1 for medium-sized compounds, comparable to typical experimental measurements. Larger compounds of interest often exhibit many conformations that can significantly differ in intramolecular interactions. Although the present capabilities allow processing even a few hundred distinct conformer structures for a given compound, many systems of interest exhibit numbers well in excess of 1000. In this study, we investigate how to reduce the number of expensive LCCSD(T) calculations for large conformer ensembles while controlling the error of the approximation. The best strategy found was to correct the results of the lower-level, surrogate model (density functional theory, DFT) in a systematic manner. It was also found that the error in the conformational contribution introduced by a surrogate model is mainly driven by a systematic (bias) rather than a random component of the DFT energy deviation from the LCCSD(T) target. This distinction is usually overlooked in DFT benchmarking studies. As a result of this work, the enthalpies of formation for 20 cannabinoid and cannabinoid-related compounds were obtained. Comprehensive uncertainty analysis suggests that the expanded uncertainties of the obtained values are below 4 kJ·mol^-1.