Ichigaku Takigawa

Catalytic N2O decomposition in the presence of O2 is a key process for addressing environmental challenges, such as greenhouse gas emissions and ozone layer depletion. However, the identification of efficient catalysts for this reaction remains challenging owing to the limitations of conventional methods. In this study, we employed a machine learning (ML) approach designed to accelerate the discovery of effective direct catalytic N2O decomposition catalysts. Starting with 51 catalysts and conducting 37 cycles of a closed-loop discovery system (ML prediction + experiment), 633 catalysts were experimentally tested. Over 10 multi-elemental catalysts exhibiting superior activity were identified, surpassing the performance of the originally identified best catalyst. Among them, Rh(1)–Pd(2)/ZrO2_EP exhibited the highest catalytic performance for N2O decomposition. Through control experiments and a combination of ex situ and in situ characterizations, we identified the essential role of each component within the catalyst system.

Data-driven methodologies have transformed the discovery and prediction of hard materials with well-defined atomic structures by leveraging standardized datasets, enabling accurate property predictions and facilitating efficient exploration of design spaces^1,2,3. However, their application to soft materials remains challenging because of complex, multiscale structure–property relationships^4,5,6. Here we present a data-driven approach that integrates data mining, experimentation and machine learning to design high-performance adhesive hydrogels from scratch, tailored for demanding underwater environments. By leveraging protein databases, we developed a descriptor strategy to statistically replicate protein sequence patterns in polymer strands by ideal random copolymerization, enabling targeted hydrogel design and dataset construction. Using machine learning, we optimized hydrogel formulations from an initial dataset of 180 bioinspired hydrogels, achieving remarkable improvements in adhesive strength, with a maximum value exceeding 1 MPa. These super-adhesive hydrogels hold immense potential across diverse applications, from biomedical engineering to deep-sea exploration, marking a notable advancement in data-driven innovation for soft materials.

The conversion of CO2 into methanol through hydrogenation represents a promising approach for CO2 utilization and sustainable chemical production. However, current industrial methods rely on copper-based catalysts, which exhibit low CO2 conversion, limited methanol yields, and require high temperatures and pressures. In this study, we utilized a machine-learning (ML) approach to develop low-temperature CO2 hydrogenation catalysts. By employing iterative ML model predictions and experimental validation in batch reactors, we screened 580 distinct catalysts and identified 33 catalysts that outperformed the previously reported highly active catalyst (Pt(3)/Mo(20)/TiO2). The best catalyst, Pt(5)/Mo(8)–Re(1)–W(0.7)/TiO2, exhibited a methanol production rate of 1.46 mmol g-¹ h-¹ in a batch reactor and a high production rate of 1.8 mmol g-¹ h-¹ in a flow reactor at 150 °C under 4 MPa (H₂/CO₂ = 3). In situ/operando spectroscopic analysis was conducted to elucidate the function of each catalyst component in methanol synthesis. Detailed analysis revealed that in the best catalyst, Pt primarily facilitated H2 dissociation, partially reduced Mo oxides were crucial in generating oxygenated species, and the presence of acidic W promoted methanol desorption from the catalyst surface. The overall methanol formation was accelerated in the presence of Re.

Selective catalytic reduction of NO_x with hydrogen (H₂-SCR) in the presence of oxygen is an environmentally friendly and sustainable emission control technology that has attracted considerable attention. However, even the most promising currently available catalysts are not sufficiently active to effectively promote this reaction, particularly at low temperatures (< 150°C). Therefore, there is an urgent need for the development of highly active H₂-SCR catalysts. Although data-science approaches, including machine learning (ML), have been suggested to accelerate the development of catalysts for such important processes, the discovery of unique catalysts using ML remains limited. This limitation stems from a common criticism of ML, namely, its perceived inability to extrapolate and identify extraordinary materials. Herein, we present an extrapolative ML approach for the development of new multi-elemental H₂-SCR catalysts. Starting with 45 catalysts as the initial dataset, we employed a closed-loop discovery system that combined ML predictions and experimental validation over 24 iterative cycles. This process enabled the experimental testing of 425 catalysts, and the ultimate identification of several catalysts with superior activity (average N₂ yield, %) over previously reported high-performance catalysts in the temperature range of 50–150°C. The optimal catalyst was found to be Pt(1.3)-Ir(0.2)/Ba(1.5)-Co(1)/H-ZSM-5(11). Notably, Co was absent from the original dataset, and the optimal catalyst composition could not be predicted by human experts.

Data-driven methodologies have revolutionized the discovery and prediction of new hard materials, such as crystal structures and high-entropy alloys1-5. However, their application to soft materials remains challenging due to the inherent complexity of their structure–property relationships6-8. Here, we present a comprehensive data-driven approach that integrates data mining, experimentation, and machine learning to develop high-performance adhesive hydrogels from scratch, tailored for demanding underwater environments. By leveraging protein databases, we devised a descriptor strategy to statistically replicate protein sequence patterns via ideal random copolymerization, enabling targeted hydrogel design and dataset construction. Using machine learning, we optimized hydrogel formulations even with a small dataset, achieving unprecedented adhesive performance. These super-adhesive hydrogels demonstrate immense potential across diverse applications, from biomedical engineering to deep-sea exploration, marking a significant advancement in the data-driven innovation for soft materials.

Manipulating the three-dimensional (3D) structures of cells is important for facilitating to repair or regenerate tissues. A self-assembly system of cells with cellulose nanofibers (CNFs) and concentrated polymer brushes (CPBs) has been developed to fabricate various cell 3D structures. To further generate tissues at an implantable level, it is necessary to carry out a large number of experiments using different cell culture conditions and material properties; however this is practically intractable. To address this issue, we present a graph-neural network-based simulator (GNS) that can be trained by using assembly process images to predict the assembly status of future time steps. A total of 24 (25 steps) time-series images were recorded (four repeats for each of six different conditions), and each image was transformed into a graph by regarding the cells as nodes and the connecting neighboring cells as edges. Using the obtained data, the performances of the GNS were examined under three scenarios (i.e., changing a pair of the training and testing data) to verify the possibility of using the GNS as a predictor for further time steps. It was confirmed that the GNS could reasonably reproduce the assembly process, even under the toughest scenario, in which the experimental conditions differed between the training and testing data. Practically, this means that the GNS trained by the first 24 h images could predict the cell types obtained 3 weeks later. This result could reduce the number of experiments required to find the optimal conditions for generating cells with desired 3D structures. Ultimately, our approach could accelerate progress in regenerative medicine.

Manipulating the three-dimensional (3D) structures of cells is important for facilitating to repair or regenerate tissues. A self-assembly system of cells with cellulose nanofibers (CNFs) and concentrated polymer brushes (CPBs) has been developed to fabricate various cell 3D structures. To further generate tissues at an implantable level, it is necessary to carry out a large number of experiments using different cell culture conditions and material properties; however this is practically intractable. To address this issue, we present a graph neural network-based simulator (GNS) that can be trained by using assembly process images to predict the assembly status of future time steps. A total of 24 (25 steps) time-series images were recorded (four repeats for each of six different conditions), and each image was transformed into a graph by regarding the cells as nodes and the connecting neighboring cells as edges. Using the obtained data, the performances of the GNS were examined under three scenarios (i.e. changing a pair of the training and testing data) to verify the possibility of using the GNS as a predictor for further time steps. It was confirmed that the GNS could reasonably reproduce the assembly process, even under the toughest scenario, in which the experimental conditions differed between the training and testing data. Practically, this means that the GNS trained by the first 24 hour images could predict the cell types obtained three weeks later. This result could reduce the number of experiments required to find the optimal conditions for generating cells with desired 3D structures. Ultimately, our approach could accelerate progress in regenerative medicine.

Designing novel catalysts is key to solving many energy and environmental challenges. Despite the promise that data science approaches, including machine learning (ML), can accelerate the development of catalysts, truly novel catalysts have rarely been discovered through ML approaches because of one of its most common limitations and criticisms—the assumed inability to extrapolate and identify extraordinary materials. Herein, we demonstrate an extrapolative ML approach to develop new multi-elemental reverse water-gas shift catalysts. Using 45 catalysts as the initial data points and performing 44 cycles of the closed loop discovery system (ML prediction + experiment), we experimentally tested a total of 300 catalysts and identified more than 100 catalysts with superior activity compared to those of the previously reported high-performance catalysts. The composition of the optimal catalyst discovered was Pt(3)/Rb(1)-Ba(1)-Mo(0.6)-Nb(0.2)/TiO₂. Notably, niobium (Nb) was not included in the original dataset, and the catalyst composition identified was not predictable even by human experts.