Search

Nanoparticles exhibit diverse structural and morphological features that are often interconnected, making the correlation of structure/property relationships challenging. In this study a multi-structure/single-property relationship of silver nanoparticles is developed for the energy of Fermi level, which can be tuned to improve the transfer of electrons in a variety of applications. By combining different machine learning analytical algorithms, including k-mean, logistic regression, and random forest with electronic structure simulations, we find that the degree of twinning (characterized by the fraction of hexagonal closed packed atoms) and the population of the {111} facet (characterized by a surface coordination number of nine) are strongly correlated to the Fermi energy of silver nanoparticles. A concise three layer artificial neural network together with principal component analysis is built to predict this property, with reduced geometrical, structural, and topological features, making the method ideal for efficient and accurate high-throughput screening of large-scale virtual nanoparticle libraries and the creation of single-structure/single-property, multi-structure/single-property, and single-structure/multi-property relationships in the near future.

Current benchmarking methods in quantum chemistry rely on databases that are built using a chemist's intuition. It is not fully understood how diverse or representative these databases truly are. Multivariate statistical techniques like archetypal analysis and K-means clustering have previously been used to summarize large sets of nanoparticles however molecules are more diverse and not as easily characterized by descriptors. In this work, we compare three sets of descriptors based on the one-, two-, and three-dimensional structure of a molecule. Using data from the NIST Computational Chemistry Comparison and Benchmark Database and machine learning techniques, we demonstrate the functional relationship between these structural descriptors and the electronic energy of molecules. Archetypes and prototypes found with topological or Coulomb matrix descriptors can be used to identify smaller, statistically significant test sets that better capture the diversity of chemical space. We apply this same method to find a diverse subset of organic molecules to demonstrate how the methods can easily be reapplied to individual research projects. Finally, we use our bias-free test sets to assess the performance of density functional theory and quantum Monte Carlo methods.

The possibility of band gap engineering in graphene opens countless new opportunities for application in nanoelectronics. In this work, the energy gaps of 622 computationally optimized graphene nanoflakes were mapped to topological autocorrelation vectors using machine learning techniques. Machine learning modeling revealed that the most relevant correlations appear at topological distances in the range of 1 to 42 with prediction accuracy higher than 80%. The data-driven model can statistically discriminate between graphene nanoflakes with different energy gaps on the basis of their molecular topology.

The design of aluminium alloys often encounters a trade-off between strength and ductility, making it challenging to achieve desired properties. Adding to this challenge is the broad range of alloying elements, their varying concentrations, and the different processing conditions (features) available for alloy production. Traditionally, the inverse design of alloys using machine learning involves combining a trained regression model for the prediction of properties with a multi-objective genetic algorithm to search for optimal features. This paper presents an enhancement in this approach by integrating data-driven classes to train class-specific regressors. These models are then used individually with genetic algorithms to search for alloys with high strength and elongation. The results demonstrate that this improved workflow can surpass traditional class-agnostic optimisation in predicting alloys with higher tensile strength and elongation. [ABSTRACT FROM AUTHOR]

The traditional design process for aluminium alloys has primarily relied upon iterative alloy production and testing, which can be time intensive and expensive. Machine learning has recently been demonstrated to have promise in predicting alloy properties based on the inputs of alloy composition and alloy processing conditions. In the search for optimal alloy concentrations that meet desired properties, as the search space expands, the optimisation process can become more time intensive and computationally expensive, depending on the methodology used. We propose a faster workflow for inverse alloy design by using multi-target machine-learning models. We train a random forest regressor to predict the concentration of alloying elements and a random forest classifier to determine the processing condition. We further analysed the inverse model and validated findings against alloys reported in the literature. [ABSTRACT FROM AUTHOR]

Machine learning can extract complex structure/property relationships but is often insufficient to explain how to control or tune the properties of materials, particularly when they are multi‐functional. This study demonstrates the value of combining multi‐target regression and multi‐target causal graphs to address the need to simultaneously control multiple properties of nanomaterials, and the need to translate these relationships into actionable insights. Using nanodiamonds as an exemplar, recursive feature elimination is first used to identify nine structural features that allow simultaneous prediction of their electron charge transfer properties and thermochemical stability to high accuracy by an interpretable random forest regressor. A multi‐target Bayesian network with domain knowledge incorporated via interactive learning using a hill‐climbing algorithm then determines how these important structural features of nanodiamonds relate to their functional properties, proposing causal paths that can be used to inform experimental design. [ABSTRACT FROM AUTHOR]

Although it has been well established that the stability and properties of graphene oxide nanostructure are strongly influenced by the concentration, type, and distribution of oxygen groups on the surface, there has yet to be a definitive way of predicting the thermochemical stability in advance of detailed and time‐consuming experimentation or simulation. In this study, a data set of over 60 000 unique graphene oxide nanoflakes and supervised machine learning methods are used to predict the probability of observation (stability) with perfect accuracy, based on a limited set of structural features that can be controlled in advance. A decision tree is used to show how the features determine the stability, and a neural network provides an equation to predict the thermodynamic stability of virtually any configuration in minutes. This enables researchers to use machine learning as research planning tool or to assist in analyzing results from microanalysis. [ABSTRACT FROM AUTHOR]

In this study a new approach to inverse design is presented that draws on the multi‐functionality of nanomaterials and uses sets of properties to predict a unique nanoparticle structure. This approach involves multi‐target regression and uses a precursory forward structure/property prediction to focus the model on the most important characteristics before inverting the problem and simultaneously predicting multiple structural features of a single nanoparticle. The workflow is general, as demonstrated on two nanoparticle data sets, and can rapidly predict property/structure relationships to guide further research and development without the need for additional optimization or high‐throughput sampling. [ABSTRACT FROM AUTHOR]

The development of interpretable structure/property relationships is a cornerstone of nanoscience, but can be challenging when the structural diversity and complexity exceeds our ability to characterise it. This is often the case for imperfect, disordered and amorphous nanoparticles, where even the nomenclature can be unspecific. Disordered platinum nanoparticles have exhibited superior performance for some reactions, which makes a systematic way of describing them highly desirable. In this study we have used a diverse set of disorder platinum nanoparticles and machine learning to identify the pure and representative structures based on their similarity in 121 dimensions. We identify two prototypes that are representative of separable classes, and seven archetypes that are the pure structures on the convex hull with which all other possibilities can be described. Together these nine nanoparticles can explain all of the variance in the set, and can be described as either single crystal, twinned, spherical or branched; with or without roughened surfaces. This forms a robust sub-set of platinum nanoparticle upon which to base further work, and provides a theoretical basis for discussing structure/property relationships of platinum nanoparticles that are not geometrically ideal. [ABSTRACT FROM AUTHOR]

The binding energy of small molecules on two‐dimensional (2D) single atom catalysts influences their reaction efficiency and suitability for different applications. In this study, the binding energy on single metal atoms to N‐doped graphene defects was predicted using random forest regression based on approximately 1700 previously generated density functional theory simulations of catalytic reactions. Three different structural feature groups containing hundreds of individual structural features were created and used to characterise the active sites. This approach was found to be accurate and reliable using either fully relaxed output structures or pre‐simulation input structures, with coefficients of determination of R2 =0.952 and R2 =0.865, respectively. The ability to predict optimal 2D‐catalysts before undertaking expensive quantum chemical calculations is an attractive basis for future research, and could be extended to other 2D‐materials. [ABSTRACT FROM AUTHOR]

The problem of feature extraction, in crystalline solid‐state systems with point defects, is considered. Novel methods for creating features for use in machine‐learning‐based predictive modeling of such systems are developed. The methods are illustrated in a case study where machine learning methods are used to predict the onset of amorphization in crystalline systems containing vacancy defects. How the methods developed may be generalized to study other problems in solid‐state materials is also discussed. [ABSTRACT FROM AUTHOR]

[Display omitted] The widespread use of aluminium alloys in the aerospace, transport and marine industries is attributed to their desirable physical properties. The relationship between the alloy composition and microstructure and the resultant mechanical properties is complicated. Machine learning (ML) has become a valuable asset in designing new alloys. The accuracy of previously utilised ML models has been increased by partitioning the alloy data set and training regressors for individual partitions. This study uses a recently reported data-driven partitioning scheme that divides the data into classes based on feature similarity. Individual regressors were trained on each class and compared with the regressor trained on the entire data set. It was revealed that individual class-based regressors are more interpretable without loss in prediction accuracy. The results indicate that the data-driven partitioning scheme outperforms traditional domain knowledge based partitioning, providing both increased model accuracy and increased model interpretability. [ABSTRACT FROM AUTHOR]

The economic and social future of nanotechnology depends on our ability and manufacture nanomaterials that avoid potential toxicity, by identifying them before they are made, used and released into the environment. Safety-by-design is a framework for including these issues at an early stage of the development process, but balancing multiple nanoparticle properties and selection criteria remains challenging. Based on a synthetic data set of over 19,000 possible sunscreen product specifications, we have used multi-target machine learning to predict the corresponding size, shape, concentration and polytype of titania nanoparticle additives. The study considers the optical properties responsible for the sun protection factor and product transparency, including the extinction coefficients for ultra violet and visible light, and the potential for toxicity due to the generation of reactive oxygen species from the photocatalytically active facets of both anatase and rutile nanoparticles, as a function of the size and shape. We predict a number of conventional forward structure/property and property/product relationships, but show that a direct structure/product relationship provides superior performance when predicting multiple properties or product specifications simultaneously. These models are then inverted, re-optimized and re-trained to provide focused, high performing inverse design models that do not require additional optimization, and are capable of identifying nanoparticle configurations outside of the training set. The ability to directly predict suitable nanoparticle structures that conform to prerequisite sun protection, transparently and potential toxicity thresholds represents a new approach to safety-by-design that can be applied to other products and materials where multiple design criteria must be met at the same time. [Display omitted] • Using multi-target random forest regression, the the performance of sunscreens can be directly predicted based on the type of TiO 2 nanoparticle additives. • Using inverse design, the type of TiO 2 nanoparticle additives is selected based on the desired SPF, aesthetic appearance and potential toxicity. • Difficult property/product relationships can be avoided by predicting structure/product relationships directly. [ABSTRACT FROM AUTHOR]

Metal-organic frameworks (MOFs) are nanoporous materials with exceptional host-guest properties poised for groundbreaking innovations in gas separation applications according to high-throughput (HT) screening data. However, MOF structural libraries are nearly infinite in practice and so statistical and information technology will play a fundamental role in implementing and rationalizing MOF virtual screening. In this work, we apply k-means clustering and archetypal analysis (AA) to identify the truly significant nanoporous structures in a large library of ~82000 virtual MOFs. Quantitative structure-property relationship (QSPR) models of the theoretical CO2 and N2 uptake capacities were also developed using a calibration set of ~16000 hypothetical MOF structures derived from the prototypes and archetype frameworks. Since uptake capacities correlated poorly to the void fraction, surface area and pore size but these properties were used to build binary classifier predictors that successfully identify "high-performing" nanoporous materials in an external test set of ~65000 MOFs with accuracy higher than 94%. The accuracy of the classification decreased for MOFs with fluorine substituents. The classification models can serve as efficient filtering tools to detecting promising high-performing candidates at the early stage of virtual high-throughput screening of novel porous materials. [ABSTRACT FROM AUTHOR]

[Display omitted] • As an essential component of informatics, conventional machine learning techniques had been useful for identifying correlational relationships within complex molecular and materials data. • Comprehensive features extraction is essential to ensure the interpretability of machine learning model outcomes. • Correlations alone are insufficient to achieve rational design of molecules and materials when causal relationships of variables involved are unclear. • Cause-and-effect relationships could be extracted using causal inference to provide mechanistic understanding about how processing/structure/property variables influence each other. • Appending causal models to conventional machine learning model workflow allows more actionable research findings to be obtained, and could better facilitate reaction engineering. While the field of nanocatalysis has benefited from the application of conventional machine learning methods by leveraging the correlations between processing/structure/property variables, the outcomes from purely correlational studies lack actionability due to missing mechanistic insights. Statistical learning, particularly causal inference, can potentially provide access to more actionable insights by allowing the discovery and verification of deeply obscured causal relationships between variables, using strong correlations identified from interpretable machine learning models as starting points. Recent studies that exemplify the collaborative usage of correlational and causal analysis in catalysis are discussed, including studies potentially benefiting from this approach. Some challenges remaining in the application of inference techniques to the field are identified and suggestions of future directions are provided. [ABSTRACT FROM AUTHOR]

Based on a general definition of a cluster and the quality of a clustering result, a new method for evaluating existing clustering algorithms, or undertaking clustering, capable of predicting the number and type of clusters and outliers present in a data set, regardless of the complexity of the distribution of points, is presented. This algorithm, referred to as iterative label spreading, can recognize the characteristics expected of a successful clustering result before any clustering algorithm is applied, providing a type of hyper‐parameter optimization for clustering. The efficacy of the algorithm, and the assessment of clustering result, are both confirmed using large benchmark two dimensional synthetic data sets, and small multidimensional data describing a set of silver nanoparticles. It is shown that the method is ideal for studying noisy data with high dimensionality and high variance, typical of data captured in materials and nanoscience. [ABSTRACT FROM AUTHOR]

The experimental discovery of graphene has produced an avalanche of theoretical and computational studies to understand the behaviour of this fascinating material. However, the intrinsic relationships between nanoscale features and graphene stability, electronic properties and reactivity remains poorly investigated. In this work, we correlate the electronic properties of 622 computationally optimized graphene structures to their structural features using machine learning algorithms. Machine learning models of the electron affinity ( E A ), energy of the Fermi level ( E F ), electronic band gap ( E G ) and ionization potential ( E I ) are calibrated with structural features of 70% of the dataset describing more than 70% of cross-validation variance. Moreover, the predictions of the values of all the properties of a test set of the remaining 30% of dataset were specially accurate with a strong correlation of R 2 ∼ 0.9. Machine learning models have tremendous potential to rapidly identify hypothetical nanostructures with desired electronic properties that, considering the latest advances in graphene synthesis and functionalization, could be experimentally prepared in a near future. [ABSTRACT FROM AUTHOR]

Although the energy of the Fermi level is of critical importance to designing electrically conductive materials, heterostructures and devices, the relationship between the Fermi energy and complex structure of graphene oxide has been difficult to predict due to competing dependencies on oxygen concentration and distribution, defects and charge. In this study we have used a data set of over 60,000 unique graphene oxide nanostructures and interpretable machine learning methods to show that the principal determinant is the ionic charge, which is in itself structure-independent. From this we define three separate, highly accurate, charge-dependent structure/property relationships and show that the Fermi energy can be predicted based on the ether concentration, hydrogen passivation or size, for the neutral, anionic and cationic cases, respectively. These important features can inform experimental design, and are remarkably insensitive to minor structural variations that are difficult to control in the lab. [Display omitted] • The Fermi energy of graphene oxide depends on whether the graphene is neutral, anionic or cationic. • Graphene oxide nanoflakes cannot be separated into their charge state using machine learning. • Manual control of the charge state results in different structure/property relationships. [ABSTRACT FROM AUTHOR]