by Dr Dimitra-Danai Varsou, Maria Antoniou and Dr Antreas Afantitis

The rise of nanotechnology and its applications has led to the development of novel nanomaterials (NMs) that differ from their bulk analogues with respect to their physicochemical properties and behaviour. The use of NMs in the EU and globally is continuously increasing with various commercial segments such as cosmetics, paints, sensors, batteries, manufacturing, etc. driving the EU NMs market growth1. Regardless of their benefits, NMs, in some cases, may be harmful to animals and humans2. Therefore, more work is needed in the fields of safety-and-sustainability-by-design (SSbD) and risk assessment of NMs.

A thorough evaluation of the potential harmful effects of NMs is associated with high experimental costs, time constraints and ethical questions as it may involve animal studies. In this course, the 3R (Replacement, Reduction, and Refinement) principles were introduced to ensure high standards of animal welfare3. Thus, reduction of use or replacement of animals can be achieved, among others, via the development of computational (in silico) models (e.g., machine learning models). Such models can be used for novel NMs design and screening to prioritise favourable candidates for synthesis and experimental evaluation. Eventually, computational models may support regulatory stakeholders in the decision-making related to NMs safety. Due to the need for the computational assessment of NMs, the European Chemicals Agency (ECHA) through the European Union Observatory for Nanomaterials (EUON) commissioned a study regarding the valid in silico modelling tools and read-across approaches, including the creation of case studies on read-across for specific (types of) NMs, which was awarded to NovaMechanics Ltd. from Cyprus. The main purpose of this study was to inform on the current developments for read-across and other in silico approaches, which are alternatives to conventional NMs testing in the NMs SSbD, hazard and risk assessments. 

In total, 190 models and methodologies were identified via desk research. These include quantitative structure–activity relationship models adapted to the NM properties (nanoQSARs), grouping and read-across methods, Adverse Outcome Pathways (AOPs), Physiologically Based Pharmacokinetics (PBPK) models and simulations. The models were retrieved from peer-reviewed scientific publications and from publicly available reports and websites of international research projects. The desk research was followed by online surveys in which 36 experts in the nanoinformatics field participated. The identified models and tools were evaluated based on criteria such as the modelling and validation processes, the data availability and reporting transparency, the existence of applicability domains, the models’ dissemination as user-friendly tools, and other identified data gaps or weaknesses. Three case studies were developed and tested to illustrate how the identified in silico methodologies can assist the NMs SSbD and their hazard and risk assessment. The first case study evaluated three grouping/read-across methodologies for the assessment of the anti-microbial activity of various carbon-based NMs. The second tested seven web applications for the prediction of TiO2-based NMs’ properties. Finally, a workflow of interdependent models was built to predict the zeta potential of metal oxide NMs in water and potassium chloride solutions (Figure 1). 



Figure 1: Main findings of the study regarding the existing in silico tools for NMs SSbD, hazard and risk assessments


The majority of the identified and evaluated studies (82%) are based on reliable modelling workflows, algorithms and methodologies and are adequately validated and interpreted. The NMs unique characteristics and inherent complexity (e.g., the properties at the nanoscale, protein corona formations, agglomeration phaenomena, etc.) are incorporated in many of the identified methods. Automation and optimisation of the modelling process, as well as the incorporation of deep learning methodologies are considered in the in silico workflows development as well. Progress has been made to create general-use read-across methodologies that have been proven to produce reliable predictions even with small datasets. The development of other approaches, such as simulations, PBPK models, and AOPs revealing the NMs toxicity mechanisms and the NMs exposure pathways to various organisms can contribute to the SSbD of novel NMs. Nonetheless, two important limitations were identified during this project, the nano-data scarcity, and the lack of integration by the interested end-users. Therefore, additional efforts are needed to address them and successfully contribute to their acceptance from the stakeholders.

The literature analysis and experts’ surveys revealed a major  challenge: the nanotoxicity data barriers covering different NM types for the development of reliable computational models. Compared to conventional chemicals for which database solutions already exist (e.g., PubChem), standardised and reliable nanomaterial-related data are limited, scattered in different sources and formats, and not always accessible. Furthermore, the lack of rich metadata associated with the NMs experimentally measured or calculated properties does not allow proper data organisation and understanding of the studied systems. Therefore, as many of the assessed methodologies are data-driven, the performance and the accuracy of the generated predictive models may be limited, impeding a wide regulatory acceptance.

FAIR (Findable, Accessible, Interoperable, and Reusable) data and metadata or Open Data initiatives are important to address the data scarcity issues. It is also a step towards data integration in the development, validation, and regulatory acceptance of automated predictive workflows. In this course, universally established test guidelines, ontologies and data harmonisation protocols may contribute to the limitation of data heterogeneity and increase interoperability and reusability of the data. Most of the experts involved in the in silico methodologies development, prefer both FAIR and Open Data to develop their models showcasing a preferability towards transparency and data accessibility.

Experimental and computational scientists should be encouraged to use electronic laboratory notebook (ELN) software to document the employed methods, to store, and digitalise their results in different formats (tables, images, etc.), as well as to capture the necessary metadata linked to each experiment while the data and metadata are being produced. The use of already developed databases such as the eNanoMapper, the NanoPharos, the NanoCommons Knowledgebase, the caNanoLab, etc. can assist the systematic data collection and consistent harmonisation. The NInChI4 initiative that supports the development of machine-readable text representations of the NMs structure and properties, may also contribute to the comprehensive description of NMs. 

Apart from the data-related issues, the key challenge is the integration of the NMs in silico assessment in their SSbD and in the related industrial, regulatory or research activities. The stakeholders’ confidence in the developed models is still low, due to the limited modelling information from the developers and the lack of sufficient data to provide a good applicability domain. The use of a standardised reporting template (e.g., QMRF5, MODA6) could be a means to overcome these issues enhancing transparency. As an illustration of lack of reporting, only around 7% of the assessed studies included a standardised modelling report. Specific guidelines, such as the OECD guidance for the validation of QSAR models7, allow scientists to report all necessary information, i.e., modelling assumptions, steps, parameters, and applicability domain limits. When models are supported by a scientific publication, reviewers should help authors improve their work during the peer-review process, ensuring that the models are properly documented. 

The disseminated models should be “translated” in a user-friendly format to maximise their utility by non-experts and serve for the design of sustainable and benign NMs. The experts involved in modelling (56%, 20 out of 36) demonstrated a mentality towards providing user-friendly workflows and public scripts. Furthermore, 63% of the assessed nanoQSAR and read-across models can be implemented as user-friendly tools as the relevant scripts are available in open repositories (e.g., in GitHub). Script availability combined with the availability of deployment platforms (e.g., Enalos Cloud Platform, Jaqpot, QsarDB, nanoHUB) can lead to the development of more nanoinformatics tools. Additionally, 38 models and methodologies were identified to be already available as user-friendly tools. As demonstrated by the case studies, most of these tools are simple to use by non-informatics experts as they consist of common user-interface elements (e.g., menus, buttons, etc.). Stakeholders can observe the results in tabular format or through plots and download them for post-processing; thus, it is easy to include them in NMs SSbD, hazard or risk assessments workflows. Nonetheless, to maintain the users’ trust these tools should be accompanied with comprehensive tutorials and examples and should be properly maintained and updated as new data are produced, to extend their applicability domains and to ensure their permanence in the future. 

The lack of validation of the developed models with “real-life cases” hinders regulatory acceptance. Industries and laboratories that perform NMs risk assessment on a regular basis, should employ available in silico tools, and compare the predictions to their conventional experimental results to help improve the tools and enhance stakeholder trust in the methods. National and international institutions could launch campaigns to highlight the time and resources savings from the use of computational methods for the design and optimisation of NMs, possibly through success stories following the examples of computer-aided drug discovery (e.g., the case of halicin, which antibacterial activity was identified by deep learning8). 

Dynamic communication and collaboration between researchers, stakeholders and regulatory experts is crucial for the development of robust and meaningful descriptors and computational models. Knowledge can be transferred between interdisciplinary groups, allowing experts to contribute to different aspects of the nanoinformatics applications, to enhance their usefulness and applicability. Such synergies will contribute to the accomplishment of the “Closer-to-the-Market-Roadmap” recommendations as established by the EU NanoSafety Cluster, for the market implementation of safe NMs and nano-enabled products9

For more information, please refer to the complete report as prepared by NovaMechanics Ltd. and published by EUON here.



  1. 1. Papadiamantis, A. G. & Afantitis, A. EU 2025: Enjoying the benefits of nanotechnology and NMs. (2023)
  2. 2. Yang, W., Wang, L., Mettenbrink, E. M., Deangelis, P. L. & Wilhelm, S. Nanoparticle Toxicology. Annu. Rev. Pharmacol. Toxicol. 61, 269–289 (2021)
  3. 3. Hubrecht, R. C. & Carter, E. The 3Rs and humane experimental technique: Implementing change. Animals 9, 1–10 (2019)
  4. 4. Lynch, I. et al. Can an InChI for nano address the need for a simplified representation of complex nanomaterials across experimental and nanoinformatics studies? Nanomaterials 10, 2493 (2020)
  5. 5. Cherkasov, A. et al. QSAR modeling: Where have you been? Where are you going to? J. Med. Chem. 57, 4977–5010 (2014)
  6. 6. The European Materials Modelling Council. Data Documentation.
  7. 7. OECD & Organisation for Economic Co-operation and Development. Guidance document on the validation of (quantitative) structure-activity relationship [(Q)SAR] models. 1–154 (2007)
  8. 8. Stokes, J. M. et al. A Deep Learning Approach to Antibiotic Discovery. Cell 180, 688-702.e13 (2020)
  9. 9. Falk, A. et al. ‘Closer to the Market’ Roadmap (CTTM). NanoSafety Clust. 1–43 (2016) doi:10.5281/ZENODO.1493492


Biographies of the authors

Dimitra-Danai Varsou is a Senior Researcher at NovaMechanics. She undertook her PhD in the area of nanoinformatics from the School of Chemical Engineering of the National Technical University of Athens (Greece). Her research interests focus on the development of computational and machine learning workflows for the prediction of properties of small molecules and nanomaterials. She has also expertise in the development of automated and optimized methodologies under the read-across framework using advanced mathematical programming algorithms. She is committed to the dissemination of the predictive models as user-friendly tools to allow their use from the broader nanosafety community.


Maria Antoniou is a recent graduate from the School of Chemical Engineering at the National Technical University of Athens, having obtained an Integrated Masters' degree with focus on Computational Methods, Physical Chemistry and Computer Programming. She conducted her diploma thesis in the School’s Laboratory of Physical Chemistry, where she developed computational models for simulating electrochemical measurements. She is currently a Junior Researcher at NovaMechanics Ltd, where she obtained a strong foundation in machine learning techniques and their application in nanomaterials. She has contributed to projects focusing on predictive models for computer-aided drug design, as well as for nanomaterial toxicity, environmental impact, and risk assessment of nanoparticles.


Antreas Afantitis PhD, MBA has a strong scientific background in the field of chem/bio/nanoinformatics, modelling. His scientific work has been published in over 90 original research articles and reviews in international peer-reviewed journals (h-index is 33). As a Director at NovaMechanics Ltd, he has led several efforts for the development of computational infrastructures of scientific and technological excellence at a national and European level that have contributed decisively to the support of new scientists. He participates as a PI with NovaMechanics in >30 multi-partner & international Projects. Currently he is the Coordinator of the materials informatics projects NanoSolveIT and CompSafeNano.






To leave a comment on this Nanopinion we welcome you to join the conversation in EUON LinkedIn discussion group.


Weitere Blogeintrag