Journal of Chemical Theory and Computation最新文献

Understanding how protein structures dictate their diverse biological functions remains one of the central and enduring challenges in structural biology. The development of AlphaFold and ESMAtlas marks a significant advance in protein science, enabling the reliable prediction of protein structure directly from amino acid sequence. This advance in structure prediction underscores the need for complementary methods that can explore conformational space and enable efficient sampling of dynamic trajectories. Here, we present TSS-Pro, a conditional generative diffusion framework that enables efficient sampling of protein conformational trajectory space. TSS-Pro takes the initial frame as conditional input and generates protein conformational trajectories. It supports two sampling strategies: (1) consecutive sampling, where each trajectory segment is generated step by step by conditioning on the final frame of the previously generated segment, enabling temporally coherent propagation of structural transitions; (2) parallel sampling, where multiple trajectory branches are independently generated from initial conditions to enhance conformational diversity. We validate TSS-Pro on three representative systems of increasing complexity: alanine dipeptide, ubiquitin, and Drosophila cryptochrome (dCRY). TSS-Pro reproduces the free energy landscape of alanine dipeptide. In the case of ubiquitin, consecutive sampling with TSS-Pro overcomes local minima and uncovers distinct conformational states of the C-terminal region. For the large protein dCRY, TSS-Pro achieves high efficiency through parallel trajectory sampling, enabling conformational and dynamic exploration typically accessible only through extensive simulations. TSS-Pro paves the way for high-throughput exploration of protein trajectories and conformational landscapes for large and complex systems.

Self-assembly of ordered structures from suitably designed building blocks is a promising approach for the generation of soft materials with optimized properties for target applications. However, it is still an open question how to generally design realizable building blocks that lead to the desired phases. In this work, DNA-functionalized nanoparticles are used as model building blocks, and the bicontinuous double gyroid is chosen as the target structure. An effective multiscale search strategy is implemented to explore a large building-block design space, where free energy calculations are first used to coarse-grain our originally fine-grained model of the building blocks and then quickly evaluate the fitness of each design. These data are then fed into a machine learning algorithm that allows obtaining predictions for all candidates in our design space through an active learning loop. Successful coarse-grained designs are identified and validated through interfacial pinning calculations with the fine-grained model. This framework leads to the development of specific, experimentally relevant designs of DNA-functionalized nanoparticles that self-assemble into the target phase. The advocated methodology can be extended to other types of building blocks and target structures.

Despite its remarkable success for a wide variety of excited states, time-dependent density functional theory (TDDFT) can fail severely for specific types of systems─for example, charge-transfer excited states, inverted singlet-triplet gap (STG) molecules, core excitations, Rydberg states, and others. Variational excited-state density functional theory (eDFT) has been demonstrated to provide accuracy similar to that of TDDFT for "well-behaved" systems. However, eDFT consistently performs better for difficult excited states, where TDDFT yields significant errors or incorrect qualitative results. In this article, the energy equations of eDFT and two-state Tamm-Dancoff approximation (TDA) TDDFT are compared to investigate the origins of both the successes and failures of these two methods for three types of excitations: core excitations, inverted STG molecules, and intermolecular charge-transfer excitations. A general decomposition method is developed for singlet and triplet excitation energies computed by using eDFT and TDDFT. Using this decomposition approach, the effects of different functionals on various components of excitation energies are analyzed. The sensitivity of eDFT excitation energies to the underlying densities is also evaluated. Our analysis reveals that eDFT excitation energies include higher contributions from the Hartree and exchange-correlation kernels compared with TDDFT. However, greater kernel contributions in eDFT by themselves do not guarantee improved accuracy. The inclusion of orbital relaxation, together with greater kernel contributions, is responsible for the superior performance of eDFT over conventional TDDFT for difficult excited states.

Transport tunnels in enzymes with buried active sites are critical gatekeepers of enzymatic function, controlling substrate access, product release, and catalytic efficiency. Despite their importance, the transient nature of these tunnels makes them difficult to study using conventional simulation methods. In this study, we systematically evaluate three coarse-grained (CG) molecular dynamics approaches─Martini with Elastic network restraints, Martini with Go̅-model restraints, and SIRAH─for their ability to characterize tunnel structure and dynamics across diverse enzyme classes. Using haloalkane dehalogenase LinB and its engineered variants as model systems, we show that CG methods accurately reproduce the geometry of tunnel ensembles observed in all-atom (AA) simulations while providing notable computational speedups. The Martini-Go̅ model performed particularly well, capturing subtle mutation-induced changes in tunnel dynamics, such as the closure of a main tunnel and the de novo opening of a transient auxiliary tunnel in LinB variants. In contrast, Martini with Elastic network restraints was limited in capturing tunnel dynamics due to the structural bias introduced by the restraints. We further validated these findings across nine enzymes from the oxidoreductase, transferase, and hydrolase classes with diverse structural folds. Although all CG methods reliably identified functionally relevant tunnels and provided fairly accurate estimates of their ensemble geometry and key bottleneck residues, they differed in their ability to replicate tunnel dynamics, with tunnel occurrences and ranking showing moderate to good correspondence with AA results. This comprehensive evaluation highlights the strengths and weaknesses of CG simulations, establishing them as powerful tools for high-throughput analysis of enzyme tunnels, which enables more efficient enzyme engineering and drug design efforts targeting these critical structural features.

Alchemical free energy perturbation (FEP) has emerged as one of the most accurate computational methods for predicting drug–protein binding affinity. However, its adoption in drug discovery workflows has been limited by two significant challenges: excessive computational requirements that demand access to HPC clusters, and high technical complexity that restricts its use to a small group of experts. Here we present ALCHEMD, a fully automated open-source platform that enables relative binding free energy (FEP-RBFE) calculations on desktop workstations, achieving 10–20 drug–protein binding predictions daily on commodity GPUs. ALCHEMD addresses these challenges through an integrated approach, including (a) intelligent preprocessing with automated reference ligand selection and nonstandard residue parametrization; (b) Common Structure Mapping algorithm that leverages 3D structural information to resolve symmetric mapping ambiguities; (c) Combined-Structure FEP methodology that introduces a new thermodynamic cycle for smoother alchemical transformations; and (d) Convergence-Adaptive Roundtrip algorithm that enables automated enhanced adaptive sampling with dynamic resource allocation. In benchmark tests, ALCHEMD achieves comparable accuracy (MUE = 0.86 kcal/mol, R² = 0.60, τ = 0.56) while requiring only 29.3 ns average simulation time per ligand pair─4–8 fold faster than conventional protocols. The platform features dual graphical and command-line interfaces, broadening accessibility to the drug discovery community.

Chirality-induced orbital-angular-momentum selectivity (CIOAMS) in electron transmission and scattering processes is investigated. Polarization of the OAM of an electron traversing chiral media is first studied via electronic wavepacket propagation using the time-dependent Schrödinger equation. Next, spatial resolution of wavepackets carrying opposite OAM, following scattering from a corrugated surface is demonstrated. This suggests that OAM may play a significant role in the mechanisms underlying chirality-induced spin selectivity, measured for electrons crossing chiral media in setups involving Mott polarimetry. Our results highlight the potential to exploit CIOAMS in innovative emerging quantum technologies.

Surface hopping simulations critically depend on the accuracy and robustness of the underlying electronic structure methods. Fully correlated approaches─such as CASPT2, MRCI, L-PDFT, and MRSF-TDDFT─that account for both dynamic and static electron correlation (without implying an exact treatment of electron correlation) offer significant promise. Still, their practical application in dynamics remains limited by the computational cost and technical challenges. In this Perspective, we examine the current status of such methods by analyzing representative surface-hopping simulations of fulvene and pyrrole, two prototypical systems for photophysical and photochemical processes. These examples demonstrate that while fully correlated methods improve the description of bond rearrangements and hot ground-state dynamics, partially correlated approaches─such as ADC(2) and TDDFT─remain sufficient for photophysical excited-state relaxation. Across methods, persistent limitations, such as active-space instabilities and potential-energy discontinuities, imply the need for improved approaches. We argue that expanding the use of generalized active spaces in the short term and advancing large active space algorithms in the long term will be crucial for making high-accuracy nonadiabatic simulations broadly reliable and accessible.

Highly glycosylated proteins known as mucins are the principal components of mucus, the gel-like secretion that protects and lubricates many tissues in the human body. Molecular dynamics (MD) simulations are a useful tool to investigate the nanoscale structure and function of proteins; however, the high molecular weight of mucins makes them a challenging target for atomistic MD simulations. To enable long-time MD simulations of large mucins, we develop and validate new coarse-grained force field parameters within the MARTINI 3 framework for the glycosylated domains of salivary mucin, MUC5B. We use atomistic MD simulations of segments of the protein backbone connected to O-glycans with the CHARMM36m force field to parameterize the bonded parameters. The structural properties of MUC5B from the MD simulations with MARTINI 3, including the radius of gyration, end-to-end distance, and solvent accessible surface area, agree well with the atomistic simulations. Our MARTINI 3 parameters reproduce the bottlebrush structure of MUC5B observed in atomistic MD simulations and previous experiments. The power-law scaling of the radius of gyration with molecular weight is within the range observed in previous experiments of mucins. Accordingly, the MARTINI 3 parameters developed and validated in this study will facilitate accurate and efficient MD simulations of mucins and other glycoproteins for a variety of application areas including food science, drug delivery, and biomaterials.

We present an evaluation of linear-response (LR) TD-DFT for the calculation of excited-state absorption (ESA) oscillator strengths. Specifically, we compare this approach with the computationally more expensive quadratic-response (QR) scheme, which we previously benchmarked, J. Chem. Theory Comput. 21 (2025) 4688, and analyze the deviations between the two approaches using various metrics. Three subsets of molecules and ESA transitions are considered: (i) subset A, a data set of 21 compact molecules comprising 53 ESA transitions, (ii) subset B consisting of 9 large molecules inspired by real-life fluorescent dyes, and (iii) subset C of selected molecules in their relaxed S₁ geometries. For subset A, we identify a clear relationship between the single-reference character of the excited states involved in the ESA process and the magnitude of the QR-LR deviations. Additionally, we observe a significant correlation between the contribution of the TD-DFT de-excitation vectors, the single-reference character of one of the two excited states, and the QR-LR discrepancies. For both subsets B and C, the correlations observed in subset A are less pronounced. Nevertheless, the largest outliers consistently involve at least one state with strong de-excitation components. Finally, we propose a simple linear correction for the unrelaxed LR oscillator strengths. Overall, LR-TD-DFT tends to almost systematically overestimate ESA oscillator strengths, especially when one of the two excited states involved exhibits large de-excitation contributions.