Laura
Escorihuela
a,
Benjamí
Martorell
*a,
Robert
Rallo
b and
Alberto
Fernández
a
aDepartament d'Enginyeria Química, Universitat Rovira i Virgili, Avinguda dels Països catalans, 26, 43007 Tarragona, Spain. E-mail: benjami.martorell@urv.cat
bAdvanced Computing, Mathematics, and Data Division, Pacific Northwest National Laboratory, Richland, WA 99352, USA
First published on 7th September 2018
Metal oxide (MeO) nanoparticles (NPs) have become common in our everyday life over the past years. However, there is still an important knowledge gap regarding their toxicological effect and, in particular, how the different physical and chemical properties of MeO NPs influence their cytotoxicity and the subsequent implications for risk assessment. This work analyses the physicochemical properties of MeO NPs that have been reported as relevant for risk assessment and the experimental and theoretical methods used to obtain them. The surface, physical and chemical properties of NPs have been critically revisited to shed light on the features that can cause toxicity. Due to the large number of existing MeO NPs, in silico studies are necessary to get a good understanding of the NPs' physicochemical properties; therefore this review focuses on the state of the art computational methods used to model MeO NP toxicity: QSAR and QSTR models and their alternative approaches provide a better understanding of MeO NP biological toxicity in organisms.
Environmental significanceThe use of nanoparticulate materials has exponentially grown during this decade due to their extraordinary properties, covering a wide range of products in the opto-electronics, pharmaceutical, medical, cosmetic and textile industries. However, the risk to human health (cytotoxic, mutagenic or carcinogenic effects) for most of them is still not well established. Alternative routes for risk assessment based on in silico methods avoid the highly expensive and time-consuming toxic evaluation of nanoparticles in the laboratory. Cheminformatic tools relate physicochemical properties with the cytotoxic effects, but standardised methods are missing. Therefore, the establishment of standard properties and mathematical models to predict toxicity is necessary for a more efficient assessment of nanoparticles. |
Particles with one or more of their dimensions in the range of a few nm up to tenths of μm have different properties, effects and behaviour relative to their microscale counterparts.3 Recent studies provided more insight on the size dependence of nanoparticle4 properties and reactivity, revealing that small sized nanoparticles (NPs) have a more variable behaviour in terms of their properties than larger size NPs, which have a more constant behaviour. For example, size dependence changes in NPs below 5 nm have more influence than changes in NPs in the range of 15 to 90 nm due to the quantum size and macro-quantum tunnelling effects.5 Another relevant effect of the smaller NPs is the direct exposure in an organism via the mechanism of entering directly inside the body and dissolving and delivering the toxic metal, described as the Trojan effect. This effect is specific for nanoscale particles given the inadvertent recognition by cell receptors.6
Due to their intrinsic properties, nanomaterials (NMs) are the cornerstone of a wide range of technologically advanced applications, with metal oxide (MeO) NPs being the most used in areas such as electronics, optics, opto-electronics, pharmacy, medicine, cosmetics and textiles.1,5 NPs are becoming more and more common in many consumer products; however there is still an important knowledge gap regarding how size influences their physicochemical properties and, in turn, their toxicity.7,8 For instance, at the nanoscale, several metal oxides are toxic, whereas they do not show any significant toxicity at the microscale.9 In 2006, Nel et al. described the mechanisms used by NMs to interact with biological systems and their toxicological effects; since then, their total comprehension has not been achieved yet.6
Therefore, NMs need a specific regulation to assess their toxicity. In the EU, the REACH10 (Registration, Evaluation, Authorisation and Restriction of Chemicals) agency directive is the current regulatory framework for chemical risk assessment and management. NMs are considered as independent “chemical substances” and therefore their registration and labelling are also regulated. The EU acknowledges that the application of REACH may cause administrative burden, affect time to market and increase marginal costs of nano-enabled products and technologies. In the United States, the Environmental Protection Agency (EPA) has a special regulation for NMs, the Toxic Substances Control Act (TSCA). NMs are referred to in TSCA as chemicals at the nanoscale. Due to their increased use in a huge range of products, in 2015 the TSCA regulation was extended to include chemical substances manufactured or processed as nanoscale materials (https://www.epa.gov/reviewing-new-chemicals-under-toxic-substances-control-act-tsca/control-nanoscale-materials-under).
Nonetheless, toxicity assessment of NMs is a daunting task that involves multiple testing conditions and endpoints, and testing of different NP configurations (i.e., different combinations of core, shell and functionalization layers, etc.). In silico testing, specifically the establishment of quantitative (nano)structure–activity relationships (QNARs), nano-quantitative structure–property relationships (nano-QSPRs) or quantitative structure–toxicity relationships (QSTRs), constitutes a cost-effective approach to fill the existing gaps in nanosafety data. The establishment of nano-QSPRs and QNARs requires (i) a detailed physicochemical and biological characterization of NMs and (ii) the development of computational nano-descriptors suitable to represent the electronic, atomic and molecular structures of NMs.
The development and validation of standard protocols for the experimental and theoretical characterization of NPs is fundamental to the generation of the high-quality data required to develop reliable nano-QSPRs and QNARs. Several reference descriptions of experimental and theoretical research protocols have been published by the Organisation for the Economic Cooperation and Development (OECD).11 In addition, the Nanosafety Cluster,12 promoted by the EU commission, helps to monitor and harmonize the European activities related to the risk assessment of NMs.
Recent results have described mathematical models linking NM structure descriptors with toxicity effects. These descriptors include physical and chemical properties such as electronic band gap, or surface properties such as surface formation energy or reactive sites.8,13,14 Regarding this relationship, for example, a band gap descriptor can be used to estimate the oxidative stress of MeO NPs.15 Recent studies separate the surface modifiers of NPs from the core of a MeO for predicting cellular uptake.16 In particular, QSAR models show that diverse combinations of NP properties can be used to classify different levels of biological response for ZnO and TiO2 NPs.17
Regarding the use of theoretical testing methods, REACH promotes the use of computational methods to implement 3R (replacement, reduction and refinement) approaches aimed at reducing and ultimately avoiding animal testing. Furthermore, REACH considers a “chemical element obtained by any manufacturing process, including any impurity deriving from the process used”. Any chemical element can be classified within different levels of impurities if it provides hazardous properties. Therefore, REACH forces NP producers and importers to provide toxicological data and environmental impact assessments (e.g., environmental exposure) when the NP concentrations are lower than 0.1% in weight. The effective implementation of REACH regulations requires the development of alternative non-testing methods (e.g., QSAR) to evaluate the toxicity of nanoparticles.
This review discusses relevant physicochemical properties of metal oxide nanoparticles and their implications for in silico nanotoxicity assessment of MeO NPs based on OECD recommendations. It is focused on the description of special properties of MeO NPs, their implications for risk assessment and their use in nano-QSAR and nano-QSPR methods for toxicity prediction. Finally, potential future research directions and challenges are discussed.
Bulk oxides are stable in a well-defined solid crystallographic structure (or a few structures, in some cases) under standard conditions. However, for smaller sizes (e.g. microscopic scale), the lattice stress must be taken into account because it can affect the structural properties up to the total disappearance of the crystallographic structure at the NP limit. Accordingly, phases with low stability in bulk form can be found at the nanoscale. This structural phenomenon has been reported for TiO2, VOx, Al2O3 and MoOx oxides.6
Nanoparticle size also influences other important features of electronic and physicochemical properties such as electrical conductivity and colour. At the nanoscale, semiconducting materials become metallic and non-magnetic particles become magnetic due to quantum-size and macro-quantum tunnelling effects. From the point of view of solid-state physics, both the superposition of bulk states and the increase in the material strength may affect electronic properties such as the band gap.
This section discusses and analyses the physicochemical properties of NPs, identified by the OECD18 as relevant for risk assessment, that are affected by the size or by the special reactivity of MeO NPs.
Fig. 1 Comparison of the percentage of atoms exposed on the surface at macro and nano scales. Adapted from the book Science at the Nanoscale.19 |
Surface atoms are not completely coordinated with respect to the interior atoms, and therefore they show a higher energy and reactivity than the fully coordinated ones. The extra energy at the surface, γ, is defined as the free energy necessary to create a new unit of area:19
(1) |
(2) |
From the perspective of nanoparticle characterisation, surface charge is usually reported as the zeta potential, which includes the electric potential in the interfacial double layer and the pKa of the particle.20 As a consequence, this property is a good toxicity predictor for low-solubility NPs, and it is considered by the OECD as a relevant parameter for fate and exposure evaluation.22 Results have confirmed that the zeta potential is affected not only by suspension conditions such as pH, temperature, ionic strength and the types of ions in suspension23 but also by intrinsic particle properties such as size and concentration. Nevertheless, these data are not always well reported in the bibliography, given that sometimes the variability of the experimental process is not described in detail.24
The zeta potential is usually measured using particle size analyzers based on laser Doppler electrophoresis. The zeta potential, ξ, is calculated from the measured electrophoretic mobility, μ, through Henry's approximation:23
(3) |
(4) |
The influence of pH on the zeta potential is explained by the protonation/deprotonation of oxides and metal centres on the surface groups of the MeO NPs. MeO NPs have a positive zeta potential in acidic environments, whereas in basic environments the property has negative values.22 For example, in the case of bare TiO2 at pH 6.0, ξ increases from 6.7 to 8.2 mV as the particle concentration varies from 0.5 to 5.0 mg L−1.
Several studies also revealed that the zeta potential could depend on particle concentration.23,25 For example, as the particle concentration increases from 1.0 to 10.0 mg L−1, the zeta potential of naked TiO2 at pH 6.0 varies from 6.7 to 11.7 mV, and from 4.7 to 10.3 mV for Fe(OH)3 NPs at pH 7.5. This effect, however, could be attributed to either a real effect or to an experimental artefact. Tantra et al.26 suggested that the shift in ξ at low particle concentration of multi-walled carbon nanotubes, silica (LUDOX) and gold NPs, was due to an increase in the contribution of the signal from extraneous particulate matter. Small changes observed in zeta potential measurements at low particle concentrations indicate the adsorption of significant counterions from the solution on the particle surface. A possible counterion12 could be the OH− coming from the dissociation of water molecules, and the bicarbonate HCO3− and carbonate CO32− from the reactions, due to the presence of carbon dioxide (CO2) in solution from ambient gas. If the particle concentration is high, then the amount of adsorbed HCO3− can be neglected and the corresponding zeta potential becomes independent of the NP concentration.23 Therefore, the above results indicate that care must be taken in choosing appropriate particle concentration conditions for electrophoretic zeta potential measurements under standardised conditions of temperature and pH.
(5) |
The chemical potential of an atom on a NP surface is higher in convex surfaces (i.e., with positive curvature) than in flat surfaces. Mass transfer from a flat surface to a convex one results in an increase of chemical potential, while the opposite occurs with concave surfaces.24
As previously mentioned, the μ of a NP dictates its shape, composition, and crystallographic structure. At the nanoscale, free energy and stress can induce changes in thermodynamic stability, modifying cell parameters27 and, as a consequence, causing structural transformations in the crystal structure of the NP with respect to the bulk.5 In other words, when mechanical and structural stabilities are balanced with free surface energy, then unstable crystalline bulk structures may become stable at the nanoscale. This phenomenon has been observed for MeO NPs such as TiO2,27 VOx,5 Al2O3,5 and MoOx.28
The ionic or covalent character of metal–oxygen bonding is related to the chemical potential of the NP. Ionicity can be affected by the NP size. The increase in ionic character is inversely proportional to the size of the particle21 and has direct effects on properties such as conductivity and chemical reactivity.29,30 In addition, the toxicity of NPs is strongly related to the electrostatic potential since NP–cell interaction mechanisms depend on electrostatic forces. Positive NPs are electrostatically attracted by negative bacterial membranes, where they can be absorbed (only this electrostatic type of interaction has been found experimentally).29
On the other hand, the conductivity of light is easily obtained by measuring the reflectivity and the absorption of solid materials. Reflectivity is a size-dependent property because it is affected by the size of particles and by the wave impedance of materials. As a result, nanoparticles are good candidates for developing high-performance optoelectronic devices such as semiconductor light-emitting diodes and laser diodes when the size is decreased as in the case of ZnO.35
(αhν) = A(hν − Eg)n | (6) |
In the case of ZnO NPs, a broad band in absorption spectra is observed at around 369 nm, characteristic for pure ZnO. Bulk ZnO has a direct band gap of 3.34 eV.37 In the case of TiO2 NPs, the band gap is 3.15 eV.38
Given their chemical properties, MeO NPs are widely used as absorbents and chemical catalysts.
Most ionic MeO NPs show a weak Brønsted activity to protonate bases. Nevertheless, SiO2, GeOx and BOx are exceptions to this rule40 due to their low valence, with the strongest Brønsted acidity appearing in oxides with valences of five or higher (WO3, MoO3, V2O5 and S-containing oxides).40
Finally, the isoelectric point (IEP) of NPs depends on the strength of the acid or base character of a material. In the context of toxicity, it has been reported that the IEP of TiO2 NPs affects their antibacterial activity.46
The solubility of a compound in a solvent at a given temperature is not only a thermodynamic property of the bulk but also depends on the dimension of the compound. Usually, the solubility of NPs is estimated by using a correction of Ostwald's equation,47 known as Freundlich's equation or Ostwald–Freundlich's equation:48
(7) |
• Solubility is inversely proportional to size and independent of the surface area of the phase, following what Gibbs and Ostwald postulated.
• When size dependence on interfacial energy is taken under consideration, Ostwald's equation is correct.
• Numerical values from Ostwald's equation are similar to those from Ostwald–Freundlich's equation, showing that size dependence on solubility increases when dissolved particles have poorer “wettability” in the solution.
The use of in silico simulations provides an alternative approach to evaluate the solubility of MeO NPs. Recently, Escorihuela et al. have developed a method based on density functional tight binding models for evaluating qualitatively the solubility of NPs in aqueous solutions. This methodology was applied to ZnO NPs in water, although it can be easily transferred to other materials and solvents.50
For partially soluble MeO NPs, toxicity is also attributed to the release of free metal ions into the solution. Examples are ZnO and CuO,51 where toxicity cannot be satisfactorily explained only by the solubility of the NPs. Furthermore, the biological response can be modified depending on the kinetic effects of solubility; after 24 hours most of the metal ions and the aggregates are dissolved. That is why metal ion release is considered as one of the most important factors for toxicity assessment.52
ROS generation can be toxic both outside and inside cells. Metal ions released from NPs can enter cells and cause toxic effects such as oxidative stress, which impacts cell viability and may ultimately result in cell death. Extracellular ROS can also induce a series of oxidative stress reactions. For instance, the presence of OH radicals formed on MeO NPs due to ultraviolet (UV) radiation is a strong antibacterial mechanism26 present in a range of materials with a wide band gap such as ZnO.54
ROS generation is a problem not only in cytotoxicity but also in other fields of science. For example, MeO NPs can create defects in polymer electrolyte membranes (PEMs) used in fuel cells. The addition of non-stoichiometric ceria NPs induces effectively ROS generation during fuel cell operation, with the consequence of less operability duration and a degradation of the PEM.55
Experimental detection of ROS is done by using luminescent probes and electron spin resonance (ESR) spectroscopy. Since direct examination of the ability of NMs to generate ROS has very poor selectivity and poor photostability due to the short life of ROS, it is necessary to use spin trap molecules in both methods. For instance, dimethyl sulfoxide (DMSO) or 2′7′-dichloro-fluorescein diacetate (DCFH-DA) is used for ROS trapping. However, this method is not selective for different species since the scavengers do not selectively trap different ROS and it is difficult to distinguish, for example, the signal corresponding uniquely to OH radicals.45
A common problem in these two methods is that some MeO NPs, such as TiO2, have photocatalytic activity. Under illumination conditions, photocatalytic degradation of the spin trap may occur, giving a decrease in UV signal intensity against time. It is also difficult to obtain reproducible ROS spectra when nanoparticles do not form stable suspensions. In spite of this, ESR is a very stable and powerful method for the experimental characterisation of ROS.
Similarly, exposure and hazard are key factors that determine the risk that nanoparticles pose to the environment and humans. The physicochemical properties of MeO NPs contribute to their exposure and hazard profiles. Due to the large number of possible MeO NP combinations (e.g. size, shape, chemistry and surface modifications), performing a complete risk assessment for each nanoparticle would require significant time and resources.57 Despite the large diversity and complexity of the MeO nanoparticle, space detailed experimental studies including in vitro assays using bacteria cultures, in vivo experiments using rodents,58,59 and NP uptake mechanisms have started to provide data to elucidate relevant cytotoxicity mechanism. Burello and Worth60 proposed a theoretical model to predict the oxidative stress potential of oxide nanoparticles by comparing the redox potentials of relevant intracellular reactions with the energy structure of oxides. Horie et al.61 reviewed the cellular response of several manufactured NPs, giving special attention to MeO NPs. It was reported that MeO NPs induced an increase in the level of cell oxidation (ROS level).62 Comparing the results of experiments in human lung carcinoma A549 cells exposed to CuO, TiO2, ZnO, CuZnFe2O4, Fe3O4, and Fe2O3 NPs showed that the intracellular ROS level was highly increased in the cells exposed to CuO nanoparticles. The opposite effect was observed in cells exposed to CeO2 NPs in an oxidant stress test, induction of oxidative stress and antioxidative activity. These discrepancies are explained by the activation of antioxidative responses in cells or because the physicochemical characteristics of CeO2 NPs were different in each experiment; therefore, no standardisation in the characterisation activity of NPs evidences opposite results in assays.
Broadly speaking, toxicity (i.e., the potential of a substance to cause adverse health effects) is measured with respect to three parameters: dose, dimension and durability.15 In the case of NPs, the properties that induce specific toxicity effects and the mechanisms that mediate the adverse effects are still largely unclear.
As a consequence, many questions still remain open in terms of NP risk assessment. For instance, the UK government has requested advice from the Royal Society and the Royal Academy of Engineering to create a group of experts in nanoscience and nanotechnology. Similarly, in the USA, the National Nanotechnology Initiative was created to settle the innovations in this field. Worldwide institutions including the Organization for Economic Cooperation and Development (OECD), European Commission (REACH and Nanosafety Cluster), European Food Safety Authorisation, and Environmental Protection Agency also provide guidance for hazard assessment and risk evaluation of NPs.7 To date, the OECD has included in its Sponsorship Programme for Testing of Manufactured Nanomaterials a list of parameters, measurements, methods and endpoints which are considered as relevant for the regulation of NMs [ENV/JM/MONO(2006)19].18 Endpoints were categorized according to the state of dispersion, aggregation of NPs, size, surface area, porosity and surface reactivity. The document groups NPs into 11 categories of NMs and analyzes 24 different physicochemical test methods including physical identification, particle size distribution, shape, aspect ratios, agglomeration and aggregation, porosity, surface composition, crystal structure, and surface charge and reactivity. In addition to the above physicochemical properties, parameters for fate and exposure assessment such as zeta potential, water solubility and dustiness were also analysed. However, the only recommendations at this point are the use of harmonized methods and control of measurement conditions. In addition, it is necessary to develop guidance for developing NP descriptors, emphasizing the order of each descriptor and how to identify them in each technique, determining if the available information, even in the literature, is adequate and reliable.
In this context, computational (i.e., in silico) methods have emerged as an alternative for the evaluation of physicochemical properties of MeO NPs. In silico approaches provide the basic framework to implement intelligent testing strategies for hazard assessment of nanomaterials. The use of data-driven approaches to establish relationships between the structure of a nanoparticle and its physicochemical properties and bioactivity profile is an effective tool for the in silico hazard assessment.
When we extrapolate these methods to nanostructures, the terms nano-QSAR and QNAR (quantitative nanostructure–activity relationship) modelling are preferred.16 However, the extrapolation of these techniques to nanoscale materials is not straightforward. The main factors that hinder the development of nano-QSPRs and nano-QSARs are data scarcity, computational cost of developing descriptors representing the whole nanostructure, and the limited knowledge of the mechanisms that regulate nano-bio interactions. The most relevant features describing NPs include particle size, size distribution, crystal structure, shape, chemical composition, surface area and chemistry, and electronic properties.63
The properties of pristine nanoparticles (i.e., as synthesized) usually experience drastic changes and dynamic behaviour under exposure conditions (e.g., biological milieu). For instance, using the primary size of a nanoparticle as a predictor variable or as the basis to compute nanoparticle descriptors may be misleading or uninformative since the initial nanostructure of the pristine material could be completely different from the structure that emerges after exposure (e.g., formation of large aggregates and attachment of proteins to the NP surface). An additional confounding factor is that structurally similar nanoparticles may interact with biological systems through various mechanisms mediated by different biological receptors. As a consequence, completely different sets of properties and descriptors can be identified as the most predictive for nanoparticles with similar structure. All the above factors should be taken into account when developing nano-QSARs and nano-QSPRs.
One of the first attempts to develop a model was performed by Puzyn et al. using descriptors obtained from quantum chemistry calculations. The model was developed using a linear regression approach (eqn (8)) to predict in vitro toxicity for Escherichia coli bacteria:8
log(EC50)−1 = 2.59 − 0.50ΔHMe+ | (8) |
Using a similar approach, Gajewicz et al. developed a nano-QSAR based on multilinear regression to predict the toxicity of metal oxides:63
log(LC50)−1 = 2.47(±0.05) + 0.24(±0.05)ΔHfc + 0.39(±0.05)Xc | (9) |
Mikolajczyk et al.64 implemented a nano-QSPR model to predict the zeta potential using the correlation of two molecular descriptors: the spherical size, φ, and the weighted energy of the highest occupied molecular orbital (HOMO) for 15 MeO NPs:
ζ = −11.26 − 4.46φ − 2.39εHOMO/nMe | (10) |
The model allowed one to quantitatively establish a relation between the zeta potential and the structural and electronic parameters of MeO NPs.
Recent development of new models for the simultaneous prediction of multiple toxicity endpoints gains importance as a comprehensive safety assessment, given the established toxicity of some NPs to both humans and the environment.65 The application of perturbation theory goes one step beyond traditional QSAR approaches. Kleandrova et al.66 used this tool on metal and metal oxide NPs. The aim was to make a QSAR model sensitive to modifications in NP compositions and experimental conditions. The descriptors used were molar volume, electronegativity, polarizability and NP size; in the case where a NM was formed by more than one element, these properties were normalized as the sum of the properties divided by the total number of atoms. Perturbation models aim to capture the sensitivity in the composition modification of the NPs versus experimental conditions and to determine if the variations in the structure and/or composition of NMs affect their toxicity. For these reasons, the original descriptors are modified by the moving average approach in order to create a new set of descriptors. Secondly, a set of pairs of NPs was created randomly wherein in each pair one particle was taken as a reference or initial state and the other particle was the predicted one. Perturbation theory was also applied by Luan et al.,67 together with quantitative structure–toxicity relationship (QSTR) in metal and metal oxide NPs, to predict different cytotoxicity profiles considering changes in sizes and measurement conditions.
A more detailed review of existing nano-QSPR and nano-QSARs can be found elsewhere.68
Due to the large number of possible MeO NPs, an exhaustive experimental evaluation would take many years to complete and would require significant economic resources. In silico modelling of NMs is a clear alternative for obtaining, from computational chemistry calculations, descriptors that can be related to the properties and toxicity of NPs using QSAR, QSPR or QSTR methodologies, of which we have discussed a few examples here.
This review provides a detailed discussion on the physicochemical properties that are considered to be the most relevant factors in the toxicity of NPs. Chemical properties and surface-related properties are the most crucial effects to describe the toxicity of NPs. Of particular interest are properties such as surface charge and pH-related effects as well as the solubility of nanoparticles and the subsequent shedding of metal ions that result in ROS formation on the NP surface.
Novel approaches to nano-QSAR/QSPR development are also needed. For instance, a specific area that requires further exploration is the development of probabilistic nano-QSAR/QSPR models based on probability density functions of descriptor values that match the size distribution of nanoparticle dispersions. Similarly, the integrated use of multiple modalities of data (e.g., image data, numeric data and text data), both for the material and for the biological entities, could potentially open the door to model performance improvement.
In addition to assessing model performance, uncertainty quantification (UQ) and uncertainty propagation techniques should be included into the data-driven modelling workflows. Implementing proper uncertainty characterization schemes will contribute to informed decision-making during nanoparticle risk assessment.
As the amount of available data increases, data-intensive approaches such as deep learning can contribute to provide better linking between the structure of a nanoparticle and its physicochemical properties and activity. One of the main advantages of deep learning is the ability to learn highly efficient data representations, making the process of feature selection unnecessary. In this context, the use of advanced techniques such as transfer learning should be explored as a potential way of leveraging existing data for bulk chemicals to bootstrap deep learning networks for nanotoxicity.
Regarding future directions for the toxicity evaluation, there is also an urgent necessity for the standardisation of the experimental tests to challenge and validate the results of modelling studies. It is necessary that this standardisation works together with computational and experimental methodologies to gain future insight into risk assessment and other challenges. The use of in silico methods and advanced machine learning techniques to obtain NP descriptors will provide in the future the next generation of basic tools to implement intelligent testing strategies for and to support the estimation of the toxicity risk assessment of nanoparticles and regulatory decision-making.
This journal is © The Royal Society of Chemistry 2018 |