RIKEN Center for Computational Science

This work presents a comprehensive theoretical investigation of magnesium-to-nitrogen (Mg···N) noncovalent interactions in substituted pyridine-MgH2 (X-Pyr.MgH2) complexes and their implications for single-molecule electron transport. Geometry optimization, binding-energy analysis, AIM, NBO, and SAPT calculations reveal electrostatically dominated Mg···N interactions with strengths ranging from -100.03 to -77.29 kJ/mol, modulated systematically by the substituent-dependent basicity of the pyridine ring. Complex formation induces measurable structural perturbations and a pronounced reduction (1-2.5 eV) in the HOMO-LUMO energy gap. DFT-NEGF simulations of Au-molecule-Au junctions show that Mg···N bonding significantly alters transmission characteristics, producing distinct quantum interference features and a substituent-dependent suppression of current near the Fermi level. The resulting I-V responses exhibit stepwise "Coulomb staircase'' behaviour, indicating quantized charge transport across the junction. These results establish Mg···N noncovalent interactions as tunable electronic perturbations that can modulate conductance in pyridine-based single-molecule junctions, providing a feasible molecular framework for exploring single-electron transistor-like behaviour.

Economic shocks in international systems propagate not only through financial channels but also through real-sector interactions, creating feedback effects that can amplify systemic risk across countries. However, country-level systemic risk assessments often rely on single-layer analyses, potentially overlooking such cross-channel dynamics. To investigate how country-level systemic risk interpretations differ across propagation layers, we constructed a multilayer network that integrates cross-border financial exposures and real-sector trade linkages. Using BIS Locational Banking Statistics and UN Comtrade data for 20 countries from 2000 to 2023, we developed a multilayer contagion framework that combines continuous within-layer propagation based on DebtRank with a threshold-based mechanism that activates cross-layer contagion when critical loss levels are exceeded. Initial shocks were calibrated using sovereign credit default swap (CDS), which implies default probabilities, to reflect market-based credit risk conditions. The results show that countries’ systemic roles and risk transmission patterns vary across layers and over time, and that incorporating cross-layer amplification reveals vulnerabilities not captured by single-layer models.

Accurate prediction of compound-protein interactions (CPIs) is crucial for chemical biology and drug discovery. Despite recent advancements, existing deep learning (DL)-based CPI models often struggle to simultaneously achieve high generalization performance, quantify prediction confidence, and ensure explainability. Here, we propose ChemGLaM, a chemical genomics language model designed to address these three crucial challenges, thereby enabling reliable and explainable CPI predictions. ChemGLaM integrates independently pre-trained chemical and protein language models through an interaction block with a cross-attention mechanism, achieving near state-of-the-art performance in predicting novel CPIs at a low computational cost. Incorporating uncertainty estimation and attention visualization enables ChemGLaM to enhance the success rate of virtual screening and to provide molecular insights into CPIs. To demonstrate the practical impact of ChemGLaM, we constructed a publicly available database containing large-scale CPI predictions for every possible pairing between all 20,434 human proteins and all 11,455 drugs and validated its practical applicability in a case study on amyotrophic lateral sclerosis. ChemGLaM marks an important step forward in addressing the challenges of AI-driven CPI exploration and drug discovery.Scientific ContributionThis study established a unified CPI prediction framework that simultaneously achieves high generalization performance, confidence quantification, and explainability. We leveraged this framework to create a community resource by constructing a comprehensive CPI database and demonstrated its practical utility by successfully prioritizing hit compounds and deconvoluting their targets in a phenotypic screening for amyotrophic lateral sclerosis.

We present an algebraic hp-multigrid method for high-order matrix-free methods. Algebraic multigrid methods often require information about matrix entries, which are not available in a matrix-free setting; however, when rediscretization for geometric multigrid is not available for a matrix-free method, coarsening must be constructed using information from the mesh. Leveraging only mesh adjacency information, this algorithm constructs an algebraic multigrid hierarchy without requiring geometric coarsening or explicit matrix assembly, making it well-suited for GPU‑accelerated architectures. This paper presents the implementation of the matrix-free method in the high-fidelity computational fluid dynamics framework Neko, which utilizes spectral element methods with an implicit-explicit scheme to solve the incompressible Navier-Stokes equations. We utilize an hp-multigrid approach, where the problem is first coarsened from high-order polynomials to low-order polynomials, and then the low-order system is further coarsened spatially in an matrix-free fashion using mesh adjacency information. Finally, we present numerical results from the Dardel and LUMI supercomputers that demonstrate the performance and scalability of our method as well as its applicability to real-world applications.

Sodium taurocholate cotransporting polypeptide (NTCP) is a hepatic transmembrane (TM) protein that functions both as a bile acid transporter and as a host receptor for hepatitis B and D viruses via the viral preS1 binding. The structural and mechanistical determinants for NTCP’s dual functions remain largely undefined. In this study, using comprehensive structure-guided alanine-scanning mutagenesis based on the cryo-electron microscopy structure of the preS1/NTCP complex, we identified 13, 8 and 9 NTCP amino acid residues critical for viral infection, preS1 binding, and bile acid transport, respectively. Key residues overlappingly regulating viral receptor and transporter functions were located primarily at TM1 and TM8, whereas TM5 and outer-surface NTCP loops mediated viral receptor-specific activity. In addition to 8 amino acids key to preS1 binding, 5 residues likely acted at a post-preS1 binding step of infection. We further found naturally-occurring single nucleotide polymorphism-associated F274C/S NTCP variants abolished viral receptor function, via the potential conformational changes in bile acid tunnel and outer-surface hollow, as analyzed by molecular dynamics simulations. Our domain-specific structural-functional map of NTCP defines the mechanism how NTCP’s dual functionality is separately regulated, and provides a framework for designing selective antiviral agents that preserve bile acid transport.

Kinesin motors use ATP to produce force in cells, yet the conformational changes that generate force remain uncertain. Here, we report structural and mechanistic insights into a minus-end-directed kinesin-14 that exhibits increased mechanical output – the variant motor binds microtubules more tightly and moves with faster velocity than wild type. High-resolution structures, together with molecular dynamics simulations, reveal previously unobserved transitions in the nucleotide hydrolysis cycle. ADP release, triggered by microtubule binding, is coupled to twisting of the central β-sheet and stabilization of the stalk prior to the power stroke. ATP binding induces stalk fluctuations and a swing of the neck mimic, an element analogous to the kinesin-1 neck linker, resembling neck linker docking in plus-end-directed kinesins. The power stroke, characterized by a large stalk rotation, is followed by motor detachment from microtubules. The subsequent recovery stroke occurs while the motor is bound to ADP and free Pi, accompanied by β-strand-to-loop transitions, or β-sheet melting, implying that β-sheet refolding facilitates Pi release. The observed twisting and melting identify the central β-sheet as the long-sought elastic element or spring required for motor force production. The transitions we observe in kinesin-14 may also apply to other kinesins – this remains to be tested.

High-fidelity pedestrian microsimulators are essential for evaluating crowd-management policies but are too computationally costly for rapid, interactive policy screening. We propose using neural network surrogate models that predict simulator outputs conditioned on network topology, demand, and policy parameters, enabling fast estimation of policy impacts without running the full simulator. We evaluate the surrogates on synthetic networks, measuring reconstruction accuracy on pedestrian flows and prediction error on a composite policy effectiveness score. Results show that a spatio-temporal graph-based surrogate provided the most accurate policy-impact estimates.

Accurately reproducing pedestrian behavior on railway station platforms is essential for safety management and congestion mitigation. This paper presents a digital-twin framework that integrates real video analysis and crowd simulation through Bayesian optimization. The method combines flow-level information from optical flow and person-level information from object detection and tracking, and reduces discrepancies between observations and simulation by automatically updating parameters. RAFT estimates pixel-level motion vectors, while ByteTrack provides individual positions and counts. From these data, differences in flow distributions, crowd counts, and positional matching are computed to form an objective function for parameter optimization. Using 360 frames of real footage over 100 iterations, the framework showed a clear reduction in error and achieved about one-third of ByteTrack’s MAE and RMSE, accurately reproducing real crowd density and movement patterns.

Evolutionary neural architecture search (ENAS) automates the design of neural network architectures but incurs significant computational costs during the evaluation phase of candidate neural networks in the evolutionary algorithm. To address this issue, surrogate models are widely adopted to estimate the performance of candidate architectures, thereby reducing resource consumption. However, existing surrogate models are primarily based on regression or classification. They are often hindered by interference from low-quality architectures, which limits search efficiency and final performance. To overcome these challenges, we propose surrogate-assisted neural architecture search with performance clustering (PCNAS). PCNAS introduces a contrastive learning-based performance clustering encoder (PCE). The PCE maps candidate architectures into an embedding space to form high-performance clusters while maximizing separation from low-performance architectures. This design mitigates the negative influence of poor-quality candidates. To enhance the discriminative power of the PCE, we incorporate a staged sample generation strategy to strengthen the hierarchical representation capacity of the PCE. Additionally, we introduce a training-free simple classifier to efficiently select high-performance architectures from the embedding space. PCNAS achieves high accuracy with low computational cost on NAS-Bench-101 and NAS-Bench- 201, outperforming previous approaches. Furthermore, PCNAS exhibits strong efficiency in the DARTS search space, discovering a high-performing architecture with a 2.53% error rate using only 0.5 GPU days of search cost, and on MobileNetV3 with ImageNet, achieving 78.12% top-1 accuracy using only 0.3 GPU days.