Search

Abstract How many near-neighbors does a molecule have? This fundamental question in chemistry is crucial for molecular optimization problems under the similarity principle assumption. Generative models can sample molecules from a vast chemical space but lack explicit knowledge about molecular similarity. Therefore, these models need guidance from reinforcement learning to sample a relevant similar chemical space. However, they still miss a mechanism to measure the coverage of a specific region of the chemical space. To overcome these limitations, a source-target molecular transformer model, regularized via a similarity kernel function, is proposed. Trained on a largest dataset of ≥200 billion molecular pairs, the model enforces a direct relationship between generating a target molecule and its similarity to a source molecule. Results indicate that the regularization term significantly improves the correlation between generation probability and molecular similarity, enabling exhaustive exploration of molecule near-neighborhoods.

Abstract Designing compounds with a range of desirable properties is a fundamental challenge in drug discovery. In pre-clinical early drug discovery, novel compounds are often designed based on an already existing promising starting compound through structural modifications for further property optimization. Recently, transformer-based deep learning models have been explored for the task of molecular optimization by training on pairs of similar molecules. This provides a starting point for generating similar molecules to a given input molecule, but has limited flexibility regarding user-defined property profiles. Here, we evaluate the effect of reinforcement learning on transformer-based molecular generative models. The generative model can be considered as a pre-trained model with knowledge of the chemical space close to an input compound, while reinforcement learning can be viewed as a tuning phase, steering the model towards chemical space with user-specific desirable properties. The evaluation of two distinct tasks—molecular optimization and scaffold discovery—suggest that reinforcement learning could guide the transformer-based generative model towards the generation of more compounds of interest. Additionally, the impact of pre-trained models, learning steps and learning rates are investigated. Scientific contribution Our study investigates the effect of reinforcement learning on a transformer-based generative model initially trained for generating molecules similar to starting molecules. The reinforcement learning framework is applied to facilitate multiparameter optimisation of starting molecules. This approach allows for more flexibility for optimizing user-specific property profiles and helps finding more ideas of interest.

Abstract REINVENT 4 is a modern open-source generative AI framework for the design of small molecules. The software utilizes recurrent neural networks and transformer architectures to drive molecule generation. These generators are seamlessly embedded within the general machine learning optimization algorithms, transfer learning, reinforcement learning and curriculum learning. REINVENT 4 enables and facilitates de novo design, R-group replacement, library design, linker design, scaffold hopping and molecule optimization. This contribution gives an overview of the software and describes its design. Algorithms and their applications are discussed in detail. REINVENT 4 is a command line tool which reads a user configuration in either TOML or JSON format. The aim of this release is to provide reference implementations for some of the most common algorithms in AI based molecule generation. An additional goal with the release is to create a framework for education and future innovation in AI based molecular design. The software is available from https://github.com/MolecularAI/REINVENT4 and released under the permissive Apache 2.0 license. Scientific contribution. The software provides an open–source reference implementation for generative molecular design where the software is also being used in production to support in–house drug discovery projects. The publication of the most common machine learning algorithms in one code and full documentation thereof will increase transparency of AI and foster innovation, collaboration and education.

Operators in thermal hazardous environments are more susceptible to burns during physical motion due to the potential deformation of thermal protective clothing. To study the influence of fabric deformation on heat transfer, a model of a three-layer thermal protective fabric system with stretching and compression deformation throughout the heat exposure and cooling phases was established and coupled with a skin heat transfer model to predict skin burns. Changes in the thermophysical properties of the fabric with deformation and skin tissue based on body location were considered. To verify the accuracy of the model, a modified radiant protective performance tester capable of simulating fabric deformation was applied. The relative error between the prediction results of the model and the experimental results does not exceed 4.68 %. The results demonstrate that stretching or compression deformation of the fabric increases heat transfer to the skin, and compression deformation generates a greater thermal hazard by accelerating the discharge of stored energy. Fabric deformation accelerates skin burns in all body parts, especially in the arms and legs. The findings of this research can provide a theoretical basis for understanding the frequent occurrence of skin burns associated with body motion and provide guidance for the design of thermal protective clothing to enhance the partial thermal protective performance.

Abstract Siamese networks, representing a novel class of neural networks, consist of two identical subnetworks sharing weights but receiving different inputs. Here we present a similarity-based pairing method for generating compound pairs to train Siamese neural networks for regression tasks. In comparison with the conventional exhaustive pairing, it reduces the algorithm complexity from O(n2) to O(n). It also results in a better prediction performance consistently on the three physicochemical datasets, using a multilayer perceptron with the circular fingerprint as a proof of concept. We further include into a Siamese neural network the transformer-based Chemformer, which extracts task-specific features from the simplified molecular-input line-entry system representation of compounds. Additionally, we propose a means to measure the prediction uncertainty by utilizing the variance in predictions from a set of reference compounds. Our results demonstrate that the high prediction accuracy correlates with the high confidence. Finally, we investigate implications of the similarity property principle in machine learning. Graphical Abstract

Amidst the rapid advancements in the fields of photovoltaics and optoelectronic devices, CsPbBr3 nanosheets (NSs) have emerged as a focal point of research due to their exceptional optical and electronic properties. This work explores the application potential of CsPbBr3 NSs in photonic and catalytic domains. Utilizing an optimized hot-injection method and a ZnBr2-assisted in situ passivation strategy, we successfully synthesized CsPbBr3 NSs with controlled dimensions and optical characteristics. Comprehensive characterization revealed that the nucleation environment and thickness significantly influenced the structure and optical performance of the materials. The results indicate that the optimized synthesis method enables control over the lateral dimensions of the nanoparticles, ranging from 9.1 ± 0.06 nm to 334.5 ± 4.40 nm, facilitating the tuning of the excitation wavelength from 460 nm (blue) to 510 nm (green). Further analyses involving photoresponse and electrochemical impedance spectroscopy demonstrated the substantial potential of these NSs in applications such as photocatalysis and energy conversion.

Metal halide perovskite semiconductors have emerged as promising materials for various optoelectronic applications due to their unique crystal structure and outstanding properties. Among different forms, perovskite nanowires (NWs) offer distinct advantages, including a high aspect ratio, superior crystallinity, excellent light absorption, and carrier transport properties, as well as unique anisotropic luminescence properties. Understanding the formation mechanism and structure–property relationship of perovskite NWs is crucial for exploring their potential in optoelectronic devices. In this study, we successfully synthesized all-inorganic halide perovskite NWs with high aspect ratios and an orthorhombic crystal phase using the hot-injection method with controlled reaction conditions and surface ligands. These NWs exhibit excellent optical and electrical properties. Moreover, precise control over the halogen composition through a simple anion exchange process enables the tuning of the bandgap, leading to fluorescence emission, covering a wide range of colors across the visible spectrum. Consequently, these perovskite NWs hold great potential for efficient energy conversion and catalytic applications in photoelectrocatalysis.

As a direct band gap semiconductor, perovskite has the advantages of high carrier mobility, long charge diffusion distance, high defect tolerance and low-cost solution preparation technology. Compared with traditional metal halide perovskites, which regulate energy band and luminescence by changing halogen, perovskite quantum dots (QDs) have a surface effect and quantum confinement effect. Based on the LaMer nucleation growth theory, we have synthesized CsPbBr3 QDs with high dimensional homogeneity by creating an environment rich in Br− ions based on the general thermal injection method. Moreover, the size of the quantum dots can be adjusted by simply changing the reaction temperature and the concentration of Br− ions in the system, and the blue emission of strongly confined pure CsPbBr3 perovskite is realized. Finally, optical and electrochemical tests suggested that the synthesized quantum dots have the potential to be used in the field of photocatalysis.

Abstract Molecular optimization aims to improve the drug profile of a starting molecule. It is a fundamental problem in drug discovery but challenging due to (i) the requirement of simultaneous optimization of multiple properties and (ii) the large chemical space to explore. Recently, deep learning methods have been proposed to solve this task by mimicking the chemist’s intuition in terms of matched molecular pairs (MMPs). Although MMPs is a widely used strategy by medicinal chemists, it offers limited capability in terms of exploring the space of structural modifications, therefore does not cover the complete space of solutions. Often more general transformations beyond the nature of MMPs are feasible and/or necessary, e.g. simultaneous modifications of the starting molecule at different places including the core scaffold. This study aims to provide a general methodology that offers more general structural modifications beyond MMPs. In particular, the same Transformer architecture is trained on different datasets. These datasets consist of a set of molecular pairs which reflect different types of transformations. Beyond MMP transformation, datasets reflecting general structural changes are constructed from ChEMBL based on two approaches: Tanimoto similarity (allows for multiple modifications) and scaffold matching (allows for multiple modifications but keep the scaffold constant) respectively. We investigate how the model behavior can be altered by tailoring the dataset while using the same model architecture. Our results show that the models trained on differently prepared datasets transform a given starting molecule in a way that it reflects the nature of the dataset used for training the model. These models could complement each other and unlock the capability for the chemists to pursue different options for improving a starting molecule.

Currently, as shown by large-scale research on two-dimensional materials in the field of nanoelectronics and catalysis, the construction of large-area two-dimensional materials is crucial for the development of devices and their application in photovoltaics, sensing, optoelectronics, and energy generation/storage. Here, using atmospheric-pressure chemical vapor deposition, we developed a method to regulate growth conditions according to the growth mechanism for WSe2 and MoSe2 materials. By accurately controlling the hydrogen flux within the range of 1 sccm and the distance between the precursor and the substrate, we obtained large-size films of single atomic layers with thicknesses of only about 1 nm. When growing the samples, we could not only obtain a 100 percent proportion of samples with the same shape, but the samples could also be glued into pieces of 700 μm and above in size, changing the shape and making it possible to reach the millimeter/submillimeter level visible to the naked eye. Our method is an effective method for the growth of large-area films with universal applicability.

Compared with conventional semiconductors, halide perovskite nanocrystals (NCs) have a unique crystal structure and outstanding optoelectronic properties, offering wide potential for applications in optoelectronic devices such as solar cells, photodetectors, light-emitting diodes, lasers, and displays. Rational technological design is providing vital support for the development of perovskite optoelectronics. Herein, monodisperse all-inorganic halide perovskite nanocrystals with consistent morphology and cubic crystal phase were synthesized employing a modified one-pot hot injection method to independently modulate the stoichiometric ratios of three precursors involving cesium salt, lead source, and halide. In combination with an anion exchange reaction, mixing two kinds of perovskite NCs with different halogens enables a transition from violet emission to green and finally to red emission over the entire visible region. Additionally, optical and electrochemical tests suggested that the as-synthesized halide perovskite NCs are promising for photoelectric catalysis applications.

Practical applications of CsPbX3 nanocrystals (NCs) are limited by their poor stability. The formation of a heterojunction between CsPbX3 NCs and oxides is an effective means to protect perovskite from polar solvents and other external factors. Significantly improving the stability and photocatalytic properties of the core/shell perovskite is very important for its application in photoelectric and photocatalytic technology. Here, we report the synthesis of asymmetrical CsPbBr3/TiO2 core–shell heterostructure NCs at the single-particle level by hot-injection liquid-phase synthesis and sol–gel method, where each CsPbBr3 NCs is partially covered by titanium dioxide. We tested not only the optical properties of the material but also the electrochemical impedance and photocurrent density of the material in sodium sulfate solution. It is shown that the type II arrangement is generated at the heterogeneous interface, which greatly facilitates the separation of electron–hole pairs and increases the carrier transport efficiency. Compared with CsPbBr3 NCs, CsPbBr3/TiO2 has higher photocatalytic efficiency. More crucially, due to the protection of the titanium dioxide shell, the product has higher long-term stability in humid air compared with bare CsPbBr3 NCs. The asymmetrical core–shell heterostructure prepared in this study not only improves the stability of CsPbX3 NCs but also provides some ideas for optoelectronic device applications and TiO2-based photocatalysts.

Abstract A main challenge in drug discovery is finding molecules with a desirable balance of multiple properties. Here, we focus on the task of molecular optimization, where the goal is to optimize a given starting molecule towards desirable properties. This task can be framed as a machine translation problem in natural language processing, where in our case, a molecule is translated into a molecule with optimized properties based on the SMILES representation. Typically, chemists would use their intuition to suggest chemical transformations for the starting molecule being optimized. A widely used strategy is the concept of matched molecular pairs where two molecules differ by a single transformation. We seek to capture the chemist’s intuition from matched molecular pairs using machine translation models. Specifically, the sequence-to-sequence model with attention mechanism, and the Transformer model are employed to generate molecules with desirable properties. As a proof of concept, three ADMET properties are optimized simultaneously: logD, solubility, and clearance, which are important properties of a drug. Since desirable properties often vary from project to project, the user-specified desirable property changes are incorporated into the input as an additional condition together with the starting molecules being optimized. Thus, the models can be guided to generate molecules satisfying the desirable properties. Additionally, we compare the two machine translation models based on the SMILES representation, with a graph-to-graph translation model HierG2G, which has shown the state-of-the-art performance in molecular optimization. Our results show that the Transformer can generate more molecules with desirable properties by making small modifications to the given starting molecules, which can be intuitive to chemists. A further enrichment of diverse molecules can be achieved by using an ensemble of models.

Abstract Except for providing thermal protection, thermal protective clothing (TPC) may also impose thermal hazards on human skin due to the discharge of energy storage. The dual thermal protective/thermal hazardous performance of TPC is rather complicated when TPC gets wet from internal and external sources and especially when it is exposed to conditions of pressurized hot steam. In this study, several indices were used to quantitatively investigate the influences of moisture location, moisture content within TPC, and their combined effect on the energy storage caused dual performance of TPC under exposure to steam. Results showed that both the increasing of external and internal moisture could decrease heat absorption of the skin and increase burn injury time under steam exposure through heat storage, but the positive effect of the external moisture amount was more obvious. The external moisture amount had no significant effect on heat discharge from TPC to the skin, whereas the internal moisture amount increased heat discharge. Empirical models were successfully established to predict the energy absorption of human skin and the burn injury time. Research outcomes of this study will contribute to thoroughly understanding the impact of moisture on dual performance under steam exposure and provide new insights into the development of functional textile materials.

Good solid-liquid mixing homogeneity and liquid level stability are necessary conditions for the preparation of high-quality composite materials. In this study, two rotor-stator agitators were utilized, including the cross-structure rotor-stator (CSRS) agitator and the half-cross structure rotor-stator (HCSRS) agitator. The performances of the two types of rotor-stator agitators and the conventional A200 (an axial-flow agitator) and Rushton (a radial-flow agitator) in the solid-liquid mixing operations were compared through CFD modeling, including the homogeneity, power consumption and liquid level stability. The Eulerian–Eulerian multi-fluid model coupling with the RNG k–ε turbulence model were used to simulate the granular flow and the turbulence effects. When the optimum solid-liquid mixing homogeneity was achieved in both conventional agitators, further increasing stirring speed would worsen the homogeneity significantly, while the two rotor-stator agitators still achieving good mixing homogeneity at the stirring speed of 600 rpm. The CSRS agitator attained the minimum standard deviation of particle concentration σ of 0.15, which was 42% smaller than that achieved by the A200 agitators. Moreover, the average liquid level velocity corresponding to the minimum σ obtained by the CSRS agitator was 0.31 m/s, which was less than half of those of the other three mixers.

Thermal protective clothing (TPC) may actually have a dual impact on human skin since it not only resists heat and mass transfer from thermal environment, but also imposes thermal hazard through thermal stored energy discharge. The air gap entrapped clothing is an important factor influencing heat and mass transfer properties. This study used several indices to qualitatively explore the impacts of air gap size and its position on the dual performance of protective textile composites. The theoretical heat and mass transfer across an air gap was also analyzed. The results demonstrated that with the increasing of air gap size, heat transmission through textile composites during both exposure and cooling periods decreased, and the 2nd and 3rd degree burn time increased. The thermal protective performance was more affected by an inner air gap rather than by a middle air gap. However, the thermal-hazardous performance caused by the stored energy discharge was more affected by a middle air gap. Empirical models were successfully established to predict the energy absorption of the skin throughout the test. Results from this study could provide proper guidance for researchers and designers to develop new protective materials to prevent burn injuries.

In order to reduce the nitrogen oxide emission of cement plant, the denitrification of decomposing furnace is studied in this paper. Based on Fluent software platform, the 2500 t/d new dry-process cement kiln decomposing furnace of a plant is modeled and simulated by using air fractional combustion technology combined with selective noncatalytic reduction technology. The model and simulation methods are verified by the field test. The effects of the urea injection position and ammonia-nitrogen molar ratio on NO, NH3, and denitrification efficiency are studied by numerical simulation. The results show that the optimal injection position of the urea solution is 12 m, the optimal ammonia/nitrogen molar ratio is 1.8, and the optimal injection velocity of the urea solution is 40 m/s. The simulation results obtained under the optimal parameters are as follows: NO concentration is 187.60 mg/m3, NH3 escape is 32.40 mg/m3, and denitrification efficiency is 74.75%.

The interfacial mechanics and electrical properties of SiC reinforced copper matrix composites were studied via the first principles method. The work of adhesion (Wad) and the interfacial energies were calculated to evaluate the stabilities of the SiC/Cu interfacial models. The carbon terminated (CT)-SiC/Cu interfaces were predicted to be more stable than those of the silicon terminated (ST)-SiC/Cu from the results of the Wad and interfacial energies. The interfacial electron properties of SiC/Cu were studied via charge density distribution, charge density difference, electron localized functions and partial density of the state. Covalent C–Cu bonds were formed based on the results of electron properties, which further explained the fact that the interfaces of the CT-SiC/Cu are more stable than those of the ST-SiC/Cu. The interfacial mechanics of the SiC/Cu were investigated via the interfacial fracture toughness and ultimate tensile stress, and the results indicate that both CT- and ST-SiC/Cu interfaces are hard to fracture. The ultimate tensile stress of the CT-SiC/Cu is nearly 23 GPa, which is smaller than those of the ST-SiC/Cu of 25 GPa. The strains corresponding to their ultimate tensile stresses of the CT- and ST-SiC/Cu are about 0.28 and 0.26, respectively. The higher strains of CT-SiC/Cu indicate their stronger plastic properties on the interfaces of the composites.

This study explored the application of shape memory alloy (SMA) springs in a multilayer protective fabric assembly for intelligent insulation that responded to thermal environment changes. Once the SMA spring was actuated, clothing layers were separated, creating an adjustable air gap between the adjacent fabric layers. The impacts of six different SMA arrangement modes and two different spring sizes on thermal protection against either a radiant heat exposure (12 kW/m2) or a hot surface exposure (400 °C) were investigated. The findings showed that the incorporation of SMA springs into the fabric assembly improved the thermal protection, but the extent to which the springs provided thermal protection was dependent on the arrangement mode and spring size. The effectiveness of reinforcing the protective performance using SMA springs depended on the ability of clothing layers to expand an air layer. The regression models were established to quantitatively assess the relationship between the air gap formed by SMA spring and the thermal protective performance of clothing. This study demonstrated the potential of SMA spring as a suitable material for the development of intelligent garments to provide additional thermal protection and thus reduce the number of clothing layers for transitional thermal protective clothing.