To identify lead-free perovskite compounds with high compositional flexibility, we developed a band gap prediction model using machine learning (ML). We analyzed CGCNN input features and evaluated factors contributing to band gap prediction, revealing that B site atomic features were dominant. Based on the analysis results, we designed a highly interpretable feature set that derived from compositional information and applied it to the model for band gap prediction. Furthermore, we trained feature-generation ML models to predict structural features, such as cell volume and B-X bond distance, from compositional information and added these features to a support vector regression (SVR) model. We confirmed that incorporating ML-generated structural features improved the accuracy of band gap prediction.
Noncovalent interactions such as hydrogen bonds and pi-pi interactions are involved in the stabilization of three-dimensional structures and specific molecular recognition. We have recently developed the Negative Fragmentation Approach (NFA) to evaluate noncovalent interactions quantitatively, using quantum chemical calculations. Although NFA is effective for small and medium-sized systems, its direct application to large macromolecules such as proteins demands further refinement. Here, we extended the original NFA to evaluate noncovalent interactions in proteins by combining it with the ONIOM method. The NFA-ONIOM scheme was applied to a hydrogen bond between triangle(5)-3-ketosteroid isomerase and equilenin, an analogue of the intermediate.
Atomic displacement parameters (ADPs) are crystallographic structural data that may represent atomic motion, possible static displacive disorder, and thermal vibration. Thermoelectric materials often contain rattling atoms to improve performance, and ADPs describe how atoms rattle. This paper discusses the ADPs of the Ba2 site of Ba8Ga16Ge30, a clathrate structure thermoelectric material. A molecular dynamics simulation approach using a neural network potential was applied to Ba8Ga16Ge30 models considering the disorder of Ga and Ge on the Ga/Ge cage sites. The calculated U is non-zero when extrapolated to 0 K, overestimates experimental U11 by 0.005-0.009 & Aring;2 at temperatures between 200 and 300 K, and overestimates experimental U22 by 0.010-0.014 & Aring;2 between 15 and 300 K.
The p-Al4Mn phase is a complex metallic alloy and an approximant of Al-Mn quasicrystals, characterized by vacancy-ordered layered structures. Density functional theory calculations were performed using a simplified single-component aluminum model, in which the Uchida stacking motif emerges spontaneously from the vacancy-ordered A-layer geometry. Structural optimization yields a stable configuration with interlayer distances close to the golden ratio, indicating a locally icosahedral-like coordination geometry. Electron localization function analysis reveals predominantly metallic bonding with strong local electron localization near vacancies, demonstrating that vacancy ordering and electronic localization stabilize the Uchida stacking motif in the p-Al4Mn phase. These results provide insight into how the coexistence of vacancy ordering and local electronic localization may be related to the origin of the brittle behavior observed in the p-Al4Mn phase.
The next version of ABINIT-MP, Open Version 2 Revision 12, is scheduled for release in spring 2026. We are also planning Revision 16, which will incorporate GPU acceleration for MP2 energy calculations. In parallel, we are preparing the Version 3 series to support open-shell and multireference systems. Furthermore, development has begun on FMO-X, a new implementation designed for GPU environments from the outset. This paper presents an overview of these developments.
Although all hexoses share the molecular formula C6H12O6, differences in hydroxyl group orientation lead to pronounced variations in crystal packing and hydrogen-bond networks. However, a systematic, site-resolved electronic comparison of intermolecular hydrogen bonding across hexose crystals remains limited. To clarify the electronic features of intermolecular hydrogen bonding, we investigated 13 hexoses using DV-X alpha molecular orbital calculations on cluster models constructed from single-crystal structures (CSD and in-house determinations). Hydrogen-bond strength was evaluated by the bond overlap population (BOP), and Full Interaction Maps (FIMs) analysis was performed for beta-aldohexoses. BOP analysis showed that the anomeric O1 site gives the largest values in beta-aldohexoses, whereas no distinct site preference appears in alpha-aldohexoses; in contrast, O1 shows the smallest values in ketohexoses. Combined BOP and FIMs analyses indicate that both the strength and site selectivity of intermolecular hydrogen bonding are governed by steric features arising from molecular structure. FIMs analysis further demonstrated that hydrogen-bond formation depends not only on electronic strength but also on spatial arrangement. The present findings provide a quantitative, site-resolved basis for understanding monosaccharide crystal structures and support rational design of rare-sugar-based functional materials by linking hydrogen-bond electronics with packing geometry.
Chemical Reaction Neural Network (CRNN) is a method that enables data-driven analysis of reaction mechanisms by embedding the principles of reaction kinetics into the architecture of a neural network. However, conventional CRNN-based methods required large amounts of training data for accurate prediction. In this study, we aim to enhance prediction accuracy by applying a clustering approach to the CRNN results. Reactions that were consistently predicted across multiple randomized initializations were identified, and the corresponding weights were fixed, thereby improving the overall prediction accuracy.
Carcinogenesis is initiated by genetic damage or mutations and involves multiple genes and signaling pathways. Proper chromosome segregation is essential for its suppression. In this study, we applied computational chemical methods to investigate SE Translocation (SET), a protein implicated in chromosome segregation. The analysis indicates that SET localizes to the centromere through electrostatic interactions between Glu206 of SET and lysine residues of Shugoshin 2 (Sgo2), including Lys38.
We attempted to reproduce molecular energies at the coupled-cluster level from molecular structures by constructing machine learning models that reproduce atomic environment energies, defined as the difference between the atomic energy in a molecule and that of the corresponding isolated atom. To obtain atomic energies from quantum chemical calculations, we adopted energy density analysis. Our results showed that the prediction accuracy of coupled-cluster molecular energies was comparable to that for density functional theory.
Circularly polarized luminescence (CPL) is the optical property of chiral molecules with promising applications in photonic devices. Oxaza[7]dehydrohelicene derivatives have attracted attention as potential high-performance CPL materials. However, strategies for designing high-performance derivatives remain unclear due to the complexity of excited-state characteristics. A knowledge graph (KG) is a framework that represents data in a graph-based structure and is applied for interpretable reasoning and visualization of relationships within data. In this study, we constructed a KG of oxaza[7]dehydrohelicene derivatives using quantum chemical calculations and conducted mechanics simulations to visualize and analyze the relationship between substituents and CPL properties.
