Explainable discovery of disease biomarkers: The case of ovarian cancer to illustrate the best practice in machine learning and Shapley analysis.

OBJECTIVE: Ovarian cancer is a significant health issue with lasting impacts on the community. Despite recent advances in surgical, chemotherapeutic and radiotherapeutic interventions, they have had only marginal impacts due to an inability to identify biomarkers at an early stage. Biomarker discovery is challenging, yet essential for improving drug discovery and clinical care. Machine learning (ML) techniques are invaluable for recognising complex patterns in biomarkers compared to conventional methods, yet they can lack physical insights into diagnosis. eXplainable Artificial Intelligence (XAI) is capable of providing deeper insights into the decision-making of complex ML algorithms increasing their applicability. We aim to introduce best practice for combining ML and XAI techniques for biomarker validation tasks. METHODS: We focused on classification tasks and a game theoretic approach based on Shapley values to build and evaluate models and visualise results. We described the workflow and apply the pipeline in a case study using the CDAS PLCO Ovarian Biomarkers dataset to demonstrate the potential for accuracy and utility. RESULTS: The case study results demonstrate the efficacy of the ML pipeline, its consistency, and advantages compared to conventional statistical approaches. CONCLUSION: The resulting guidelines provide a general framework for practical application of XAI in medical research that can inform clinicians and validate and explain cancer biomarkers.

The pure and representative types of disordered platinum nanoparticles from machine learning.

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.

Three-Dimensional Branched and Faceted Gold-Ruthenium Nanoparticles: Using Nanostructure to Improve Stability in Oxygen Evolution Electrocatalysis.

Achieving stability with highly active Ru nanoparticles for electrocatalysis is a major challenge for the oxygen evolution reaction. As improved stability of Ru catalysts has been shown for bulk surfaces with low-index facets, there is an opportunity to incorporate these stable facets into Ru nanoparticles. Now, a new solution synthesis is presented in which hexagonal close-packed structured Ru is grown on Au to form nanoparticles with 3D branches. Exposing low-index facets on these 3D branches creates stable reaction kinetics to achieve high activity and the highest stability observed for Ru nanoparticle oxygen evolution reaction catalysts. These design principles provide a synthetic strategy to achieve stable and active electrocatalysts.

On reverse Monte Carlo constraints and model reproduction.

Reverse Monte Carlo (RMC) simulations were performed to investigate the effectiveness of any combination of five experimentally motivated constraints on the reproduction of a test case, a ternary ab initio model. It was found that low energy structures fitting a variety of constraints commonly used in the RMC methodology could still provide an incorrect description of the chemical structural unit populations in multi-elemental systems. It is shown that the use of an elemental bond type constraint is an effective way to avoid this. © 2017 Wiley Periodicals, Inc.

Machine Learning for Silver Nanoparticle Electron Transfer Property Prediction.

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.

Machine learning and genetic algorithm prediction of energy differences between electronic calculations of graphene nanoflakes.

Computational screening is key to understanding structure-function relationships at the nanoscale but the high computational cost of accurate electronic structure calculations remains a bottleneck for the screening of large nanomaterial libraries. In this work we propose a data-driven strategy to predict accuracy differences between different levels of theory. Machine learning (ML) models are trained with structural features of graphene nanoflakes to predict the differences between electronic properties at two levels of approximation. The ML models yield an overall accuracy of 94% and 88%, for energy of the Fermi level and the band gap, respectively. This strategy represents a successful application of established ML methods to the selection of optimum level of theory, enabling more rapid and efficient screening of nanomaterials, and is extensible to other materials and computational methods.

Quantitative Structure-Property Relationship Modeling of Electronic Properties of Graphene Using Atomic Radial Distribution Function Scores.

The intrinsic relationships between nanoscale features and electronic properties of nanomaterials remain poorly investigated. In this work, electronic properties of 622 computationally optimized graphene structures were mapped to their structures using partial-least-squares regression and radial distributions function (RDF) scores. Quantitative structure-property relationship (QSPR) models were calibrated with 70% of a virtual data set of 622 passivated and nonpassivated graphenes, and we predicted the properties of the remaining 30% of the structures. The analysis of the optimum QSPR models revealed that the most relevant RDF scores appear at interatomic distances in the range of 2.0 to 10.0 Å for the energy of the Fermi level and the electron affinity, while the electronic band gap and the ionization potential correlate to RDF scores in a wider range from 3.0 to 30.0 Å. The predictions were more accurate for the energy of the Fermi level and the ionization potential, with more than 83% of explained data variance, while the electron affinity exhibits a value of ∼80% and the energy of the band gap a lower 70%. QSPR models have tremendous potential to rapidly identify hypothetical nanomaterials with desired electronic properties that could be experimentally prepared in the near future.

Catalytic potential of highly defective (211) surfaces of zinc blende ZnO.

Zinc blende (ZB) ZnO has gained increasing research interest due to its favorable properties and its stabilization on the nanoscale. While surface properties are important on the nanoscale, the studies on ZB ZnO surface properties are rare. Here we have performed first principles calculations of the energies and structures of ZB and wurtzite (WZ) ZnO surfaces. Our results indicate that, among the four surfaces parallel to the polar axes, such as (101̄0) and (112̄0) of the WZ phase and (110) and (211) of the ZB phase, the polar (211) surface has substantially lower surface vacancy formation energies than the others, which makes ZB ZnO promising for catalytic applications. Our results also imply that the stabilization of ZB ZnO on the nanoscale is due to some mechanisms other than surface energies.

Functionalized Nanodiamonds for Biological and Medical Applications.

Nanodiamond is a promising material for biological and medical applications, owning to its relatively inexpensive and large-scale synthesis, unique structure, and superior optical properties. However, most biomedical applications, such as drug delivery and bio-imaging, are dependent upon the precise control of the surfaces, and can be significantly affected by the type, distribution and stability of chemical funtionalisations of the nanodiamond surface. In this paper, recent studies on nanodiamonds and their biomedical applications by conjugating with different chemicals are reviewed, while highlighting the critical importance of surface chemical states for various applications.

Size and shape dependent deprotonation potential and proton affinity of nanodiamond.

Many important reactions in biology and medicine involve proton abstraction and transfer, and it is integral to applications such as drug delivery. Unlike electrons, which are quantum mechanically delocalized, protons are instantaneously localized on specific residues in these reactions, which can be a distinct advantage. However, the introduction of nanoparticles, such as non-toxic nanodiamonds, to this field complicates matters, as the number of possible sites increases as the inverse radius of the particle. In this paper we present > 10(4) simulations that map the size- and shape-dependence of the deprotonation potential and proton affinity of nanodiamonds in the range 1.8-2.7 nm in average diameter. We find that while the average deprotonation potential and proton affinities decrease with size, the site-specific values are inhomogeneous over the surface of the particles, exhibiting strong shape-dependence. The proton affinity is strongly facet-dependent, whereas the deprotonation potential is edge/corner-dependent, which creates a type of spatial hysteresis in the transfer of protons to and from the nanodiamond, and provides new opportunities for selective functionalization.

Activity of ZnO polar surfaces: an insight from surface energies.

The calculation of the accurate surface energies for (0001) surfaces of wurtzite ZnO is difficult because it is impossible to decouple the two inequivalent (0001)-Zn and (0001¯)-O surfaces. By using a heterojunction model we have transformed the uncertainty of the surface energies into that of interface energies which is much smaller than the former and hence estimated the surface energies to a high degree of accuracy. It is found that the oxygen terminated (0001¯)-O face of the wurtzite phase and (1¯1¯1¯O of the zinc blende phase are more stable than their Zn-terminated counterparts within the major temperature and oxygen partial pressure range accessible to experiment. The instability of Zn-terminated polar surfaces explains the experimentally observed high activity of these surfaces. The effects of native surface vacancies on the surface energies have also been discussed. These results provide insights into the modification of the surface stability and activity of ZnO nanoparticles.

Classification of battery compounds using structure-free Mendeleev encodings.

Machine learning is a valuable tool that can accelerate the discovery and design of materials occupying combinatorial chemical spaces. However, the prerequisite need for vast amounts of training data can be prohibitive when significant resources are needed to characterize or simulate candidate structures. Recent results have shown that structure-free encoding of complex materials, based entirely on chemical compositions, can overcome this impediment and perform well in unsupervised learning tasks. In this study, we extend this exploration to supervised classification, and show how structure-free encoding can accurately predict classes of material compounds for battery applications without time consuming measurement of bonding networks, lattices or densities. SCIENTIFIC CONTRIBUTION: The comprehensive evaluation of structure-free encodings of complex materials in classification tasks, including binary and multi-class separation, inclusive of three classifiers based on different logic function, is measured four metrics and learning curves. The encoding is applied to two data sets from computational and experimental sources, and the outcomes visualised using 5 approaches to confirms the suitability and superiority of Mendeleev encoding. These methods are general and accessible using source software, to provide simple, intuitive and interpretable materials informatics outcomes to accelerate materials design.

Harnessing the influence of reactive edges and defects of graphene substrates for achieving complete cycle of room-temperature molecular sensing.

Molecular doping and detection are at the forefront of graphene research, a topic of great interest in physical and materials science. Molecules adsorb strongly on graphene, leading to a change in electrical conductivity at room temperature. However, a common impediment for practical applications reported by all studies to date is the excessively slow rate of desorption of important reactive gases such as ammonia and nitrogen dioxide. Annealing at high temperatures, or exposure to strong ultraviolet light under vacuum, is employed to facilitate desorption of these gases. In this article, the molecules adsorbed on graphene nanoflakes and on chemically derived graphene-nanomesh flakes are displaced rapidly at room temperature in air by the use of gaseous polar molecules such as water and ethanol. The mechanism for desorption is proposed to arise from the electrostatic forces exerted by the polar molecules, which decouples the overlap between substrate defect states, molecule states, and graphene states near the Fermi level. Using chemiresistors prepared from water-based dispersions of single-layer graphene on mesoporous alumina membranes, the study further shows that the edges of the graphene flakes (showing p-type responses to NO2 and NH3) and the edges of graphene nanomesh structures (showing n-type responses to NO2 and NH3) have enhanced sensitivity. The measured responses towards gases are comparable to or better than those which have been obtained using devices that are more sophisticated. The higher sensitivity and rapid regeneration of the sensor at room temperature provides a clear advancement towards practical molecule detection using graphene-based materials.

Direct comparison of kinetic and thermodynamic influences on gold nanomorphology.

Under a given set of conditions, nanomaterials can crystallize into structures that are entirely inconsistent with the bulk material and may adopt a range of faceted morphologies that depend on the particle size. A size-dependent phase diagram, a graphical representation of the chemical equilibrium, offers a convenient way to describe this relationship among the size, morphology, and thermodynamic environment. Although creating such a diagram from conventional experiments is extremely challenging (and costly), theory and simulation allow us to use virtual experiments to control the temperature, pressure, size, structure and composition independently. Although the stability and morphology of gold nanoparticles has been add-ressed numerous times in recent years, a critical examination of the literature reveals a number of glaring contradictions. Typically gold nanoparticles present as multiply-twinned structures, such as icosahedra and decahedra, or faceted monocrystalline (fcc) shapes, such as truncated octahedra and cuboctahedra. All of these shapes are dominated by various fractions of {111} and {100} facets, which have different surface atom densities, electronic structure, bonding, chemical reactivities, and thermodynamic properties. Although many of the computational (and theoretical) studies agree on the energetic order of the different motifs and shapes, they do not necessarily agree with experimental observations. When discrepancies arise between experimental observations and thermodynamic modeling, they are often attributed to kinetics. But only recently could researchers analytically compare the kinetics and thermodynamics of faceted nanoparticles. In this Account, we follow a theoretical study of the size, shape, and structure of nanogold. We systematically explore why certain shapes are expected at different sizes and (more importantly) why others are actually observed. Icosahedra are only thermodynamically preferred at small sizes, but we find that they are the most frequently observed structures at larger sizes because they are kinetically stable (and coarsen more rapidly). In contrast, although the phase diagram correctly predicts that other motifs will emerge at larger sizes, it overestimates the frequency of those observations. These results suggest either a competition or collaboration between the kinetic and thermodynamic influences. We can understand this interaction between influences if we consider the change in shape and the change in size over time. We then use the outputs of the kinetic model as inputs for the thermodynamic model to plot the thermodynamic stability as a function of time. This comparison confirms that decahedra emerge through a combination of kinetics and thermodynamics, and that the fcc shapes are repressed due to an energetic penalty associated with the significant departure from the thermodynamically preferred shape. The infrequent observation of the fcc structures is governed by thermodynamics alone.

Modeling polydispersive ensembles of diamond nanoparticles.

While significant progress has been made toward production of monodispersed samples of a variety of nanoparticles, in cases such as diamond nanoparticles (nanodiamonds) a significant degree of polydispersivity persists, so scaling-up of laboratory applications to industrial levels has its challenges. In many cases, however, monodispersivity is not essential for reliable application, provided that the inevitable uncertainties are just as predictable as the functional properties. As computational methods of materials design are becoming more widespread, there is a growing need for robust methods for modeling ensembles of nanoparticles, that capture the structural complexity characteristic of real specimens. In this paper we present a simple statistical approach to modeling of ensembles of nanoparticles, and apply it to nanodiamond, based on sets of individual simulations that have been carefully selected to describe specific structural sources that are responsible for scattering of fundamental properties, and that are typically difficult to eliminate experimentally. For the purposes of demonstration we show how scattering in the Fermi energy and the electronic band gap are related to different structural variations (sources), and how these results can be combined strategically to yield statistically significant predictions of the properties of an entire ensemble of nanodiamonds, rather than merely one individual 'model' particle or a non-representative sub-set.

Diamond nanoparticles as a new platform for the sequestration of waste carbon.

The use of carbon nanostructures to capture and store waste carbon, such as methane and carbon dioxide, is intrinsically attractive, particularly if the same molecules can be subsequently used as synthetic precursors. However, to facilitate adsorption of these highly stable species high pressures are required, and fragile carbon-based nanostructures may not survive. By combining electronic structure simulations and ab initio thermodynamics, we have investigated the thermochemical conditions required to adsorb CH, CH2, CO and CO2 on diamond nanoparticles, which can withstand higher temperatures and pressures than alternative carbon-based nanostructures. We find that, while CO2 must be over-saturated to facilitate stable adsorption (with high efficiency), the strength of the resultant C-O bonds means that desorption will not occur spontaneously when atmospheric pressure is resumed.

Site-dependent stability and electronic structure of single vacancy point defects in hexagonal graphene nano-flakes.

Graphene nano-flakes and quantum dots have considerable potential as components for nanodevices, since the finite in-plane dimension and additional edge and corner states provide potential for band gap engineering. However, like semi-infinite graphene membranes, they may contain different configurations of vacancy point defects that may be difficult to predict or control. In this paper we use density functional tight binding simulations to explore the impact of different geometric configurations of vacancies in unterminated (radical), mono-hydride and di-hydride terminated nano-flakes with zigzag or armchair edges. The results reveal that the planar structure is more uniformly preserved (with less distortion) when vacancies are located near the edges and corners, due to the combined effect of vacancy-edge-corner reconstructions, and passivating the circumference reduces the scattering of the band gap, but not the scattering of the energy of the Fermi level. In general, and regardless of the possible application, the use of zigzag-edged nano-flakes with stable edge/corner passivation is desirable to ensure reliability, and reduce the impact of an unknown number and configurations of vacancies.

Unsupervised machine learning discovers classes in aluminium alloys.

Aluminium (Al) alloys are critical to many applications. Although Al alloys have been commercially widespread for over a century, their development has predominantly taken a trial-and-error approach. Furthermore, many discrete studies regarding Al alloys, often application specific, have precluded a broader consolidation of Al alloy classification. Iterative label spreading (ILS), an unsupervised machine learning approach, was used to identify the different classes of Al alloys, drawing from a specifically curated dataset of 1154 Al alloys (including alloy composition and processing conditions). Using ILS, eight classes of Al alloys were identified based on a comprehensive feature set under two descriptors. Further, a decision tree classifier was used to validate the separation of classes.

Complex Dispersion of Detonation Nanodiamond Revealed by Machine Learning Assisted Cryo-TEM and Coarse-Grained Molecular Dynamics Simulations.

Understanding the polydispersity of nanoparticles is crucial for establishing the efficacy and safety of their role as drug delivery carriers in biomedical applications. Detonation nanodiamonds (DNDs), 3-5 nm diamond nanoparticles synthesized through detonation process, have attracted great interest for drug delivery due to their colloidal stability in water and their biocompatibility. More recent studies have challenged the consensus that DNDs are monodispersed after their fabrication, with their aggregate formation poorly understood. Here, we present a novel characterization method of combining machine learning with direct cryo-transmission electron microscopy imaging to characterize the unique colloidal behavior of DNDs. Together with small-angle X-ray scattering and mesoscale simulations we show and explain the clear differences in the aggregation behavior between positively and negatively charged DNDs. Our new method can be applied to other complex particle systems, which builds essential knowledge for the safe implementation of nanoparticles in drug delivery.