The COSMOS Project: A Foundation for the Future of Computational Modelling of Repeat Dose Toxicity

Home / New Perspectives / Organ Toxicity / The COSMOS Project: A Foundation for the Future of Computational Modelling of Repeat Dose Toxicity

Organ Toxicity

The COSMOS Project: A Foundation for the Future of Computational Modelling of Repeat Dose Toxicity

Mark T.D. Cronin, Judith C. Madden, and Andrea-Nicole Richarz, School of Pharmacy and Chemistry, Liverpool John Moores University

Published: August 29, 2012

About the Author(s)
Mark Cronin co-ordinates the COSMOS project. He is Professor of Predictive Toxicology in the School of Pharmacy and Chemistry at Liverpool John Moores University, England. He has over 25 years of expertise in the use of (Q)SARs and other in silico technologies to predict environmental and human health effects.

Mark Cronin
James Parsons Building
Liverpool John Moores University
Byrom Street
Liverpool, L3 3AF
United Kingdom

Dr. Judith Madden is a Reader in Molecular Design at Liverpool John Moores University. She has worked for over 20 years in the area of predicting biological activity of xenobiotics based on their chemical structure, considering both inherent activity and pharmacokinetic influences.

Judith Madden
James Parsons Building
Liverpool John Moores University
Byrom Street
Liverpool, L3 3AF
United Kingdom

Dr. Andrea-Nicole Richarz is the Project Manager for the COSMOS Project at Liverpool John Moores University. She holds a PhD in Chemistry from the Technical University Berlin, Germany and has many years of experience with project management specialising in the field of predictive toxicology and alternatives to animal testing.

Andrea-Nicole Richarz
James Parsons Building
Liverpool John Moores University
Byrom Street
Liverpool, L3 3AF
United Kingdom

The COSMOS project (Integrated In Silico Models for the Prediction of Human Repeated Dose Toxicity of COSMetics to Optimise Safety) is rising to the challenges of computational modelling of repeat dose toxicity data. The project aims to be at the centre of efforts to integrate reliable and open source toxicity data, thresholds of toxicological concern (TTC), grouping for read-across and modelling of biokinetics with the opportunities offered by informatics and the toxicity pathway approach. It is already resolving problems often encountered with the storage, searching and retrieval of repeat dose toxicity data, as well as updating the TTC approach and supporting efforts in developing groups (categories) of molecules for read-across.

The Challenge

Prediction of repeat dose toxicity has posed a real challenge to computational modelling to provide a viable alternative to animal testing. To understand this challenge, it must be understood that toxicity resulting from long-term exposure to a chemical is a consequence of perturbing a biological system at the cellular, tissue and organ (or multiple organ) level. Several factors can influence the outcome including toxicokinetics and physiological adaptive response mechanisms. Repeat dose toxicity testing provides a No Observable (Adverse) Effect Level (NO(A)EL) which is used in quantitative risk assessment of chemicals; data from 28-day or 90-day rodent oral toxicity assays are typically used. Because of these factors, added to the fact that there are many potential mechanisms and may be multiple (interacting) organ systems involved in eliciting the toxicity, in silico models have previously been considered too simplistic to model such complex interactions.

The European Union’s Cosmetics Regulation will require cessation of all animal testing for toxicological purposes of cosmetic ingredients (marketed in the EU) in 2013. In addition, the EU’s REACH regulation will soon require assessment of lower tonnage chemicals for which fewer toxicity data are available. Hence there is an urgent need to develop robust alternatives to animal tests. This need prompted a review of the “state of the art” of non-test alternatives, published by Adler et al. (2011), which concluded that we are not yet close to reliable (in vitro or in silico) alternatives to many toxicity endpoints including repeat dose effects. Considering all of these factors, the following problems and possible solutions are evident in developing acceptable computational models for repeat dose toxicity.

  • Whilst it is generally acknowledged that repeat dose toxicity is not readily amenable to “classical” correlative computational modelling such as quantitative structure-activity relationships (QSARs), recent advances in in silico modelling, such as read-across methods, can provide a solution. It is important to challenge the preconception that computational models cannot be sufficiently sophisticated to provide useful predictions.
  • In order to develop models for repeat dose toxicity, access to high quality toxicity data is required. These need to be stored in a way that allows relevant details (such as experimental protocol, organ level effects, pathology, etc.) to be readily accessed. There is currently no single source of repeat dose toxicity data; in addition, many data may be available within industry with no means of extracting them. What is required is a comprehensive, flexible and reliable database tailored for repeat dose toxicity from which useful predictive models can be developed.
  • The Threshold of Toxicological Concern (TTC) approach identifies an exposure threshold below which no adverse effect on humans is anticipated. This approach has been adopted by the US Food and Drug Administration to determine safe levels for food contact substances. The potential to update and adapt this approach for the safety assessment of cosmetics ingredients needs to be investigated.
  • The complexity of modelling repeat dose toxicity endpoints is due, in part, to the lack of definition and (in some cases) understanding of the mechanisms of action of chronic toxicity. This has meant that in many cases it was not possible to determine the relationships that linked the chemistry behind the process to the toxic effect. This has resulted in more concerted attempts to model individual pathways which lead to effects, rather than the effect itself.
  • The modelling of repeat dose toxicity needs to utilise techniques such as category formation and read-across. Read-across is an in silico method that enables the activity of a compound of interest to be inferred using information that is known about similar compounds. A rational method is used to group chemicals together into categories of similar compounds. Data for known compounds within the category need not be restricted to a definitive toxic endpoint (e.g. ED50 value) but can include information that is indicative of toxic effect (e.g. potency in an associated in vitro assay).
  • Current initiatives, such as the Tox 21 program, are subjecting a range of compounds of interest to a vast array of in vitro tests. The results from such assays can provide useful insight into mechanisms of toxic effects. There is a need to link computational modelling to the outcome of these in vitro assays to gain better understanding of toxicity pathways in vivo.
  • One of the key factors in predicting the overall biological activity of a compound is correlating toxic effect to the dose that is received by the target organ. This can differ considerably from the dose administered to the animal due to biokinetic factors (absorption, distribution, metabolism and excretion). Traditional approaches, such as QSAR, that correlate dose administered to biological activity can produce misleading models if such factors are not accounted for. Biokinetic models are currently under-developed. Improvements are required to allow for the extrapolation of information from toxicity pathways and in vitro assays to organ level effects in humans.

Modelling must move away from being a retrospective, and often academic, analysis of data and information which has often been performed in isolation. There is now a paradigm-shift towards developing models based on an understanding of the underlying mechanisms involved in eliciting an adverse effect. Development of new models will require the integration of several approaches, such as category formation and read-across, in vitro methods, in vitro to in vivo extrapolation (IVIVE) models, etc. The aim is to develop flexible and usable models that may achieve the ultimate goal of predicting NO(A)EL values using entirely non-test alternative methods. Such an aspiration requires a multinational effort to pool expertise and resources into a unique project entirely focused on addressing this challenge.

Meeting the Challenge: The COSMOS Project

The COSMOS project is a five year project (start date 1 January 2011) funded by the European Commission and Cosmetics Europe in a unique partnership. The project is co-ordinated by Mark Cronin at Liverpool John Moores University, England and links together groups within the European Union and the US. It is part of the SEURAT-1 cluster, a 50 million euro research initiative comprising seven projects in total, with the overall aims of laying the foundation to provide robust alternatives to understanding and predicting organ level toxicity following repeated exposure to chemicals. There are close links through SEURAT-1 to other projects in the EU and, of course, the activities of Tox21 in the US. COSMOS is the only project in the SEURAT-1 cluster dedicated to computational modelling and comprises 15 partners, including universities, small and medium enterprises (SMEs), research institutes and regulatory agencies. These partners provide a unique combination of expertise giving the project the impetus to make real progress in the development and implementation of non-animal alternatives.

Initial Progress Within COSMOS

Novel applications of informatics are at the heart of much that COSMOS will offer. For instance, with an emphasis towards applications in the cosmetics industry, an inventory of cosmetics ingredients has been compiled from the European Commission’s Cosmetics Ingredients (CosIng) database and the US Personal Care Products Council (PCPC) lists. This allows, for the first time, quality assured chemical structures to be associated with cosmetics ingredients. The inventory will be incorporated into the COMSOS database and will be flexibly searchable. It has already allowed for the chemical space associated with ingredients to be defined and used to assess the current TTC approaches.

The COSMOS database (DB) itself will be a relational database allowing for the linking of chemical structure to repeat dose toxicity data. This will provide a solution to one of the key challenges in developing computational models i.e. it will deliver a single, comprehensive and reliable resource for repeat dose toxicity data. The database will be constructed using robust data curation and quality assessment strategies. It is based around the CERES database format created in partnership with Molecular Networks and donated by the US FDA (both organisations are partners in the COSMOS project). Results for repeat dose toxicity tests will be recorded in detail; the database will be sufficiently flexible to enable recording of additional toxicity results as they become available. COSMOS DB and the data within it will be made freely available. This will provide an invaluable resource for the modelling efforts as well as a platform to encourage data sharing and donation. With regard to data sharing, one interested party from industry has already provided previously unpublished data for inclusion in COSMOS DB. The format of the database is such that there is also an option for interested parties to donate data in the form of robust study reports. These data can have different levels of access, ranging from the donor only, the consortium or publicly available. When available they can be utilised for data modelling and model validation purposes. A timetable for the public release of the inventory and the database will soon be available on the project website (

Additional progress has been made in adapting the TTC approach, i.e. within COSMOS the method is being extended and expanded for application to cosmetic ingredients. This has taken a number of key directions, firstly a dataset of reliable NOEL values has been created that greatly extends that originally provided in the Munro TTC dataset (Munro et al. 1996). For example, the COSMOS database includes surfactants, which are not present in the Munro dataset. This will provide greater relevance to the determination of TTC for materials used in cosmetics. Further, the chemical space of the cosmetics inventory within the COSMOS DB has been analysed; this has demonstrated that, with some exceptions (e.g. silicon-containing compounds, molecules with long aliphatic chains, etc.), there is a good overlap with cosmetic space and that of the original Munro dataset. The relevance of the dermal route of exposure to TTC (bearing in mind its importance for cosmetics and the fact that most NOELs are from oral dosing) is being examined. An extensive database of dermal penetration data is also being curated and quality assured.

The Cramer scheme (proposed by Cramer et al. 1978) is the most commonly used scheme to classify chemicals for TTC for non-cancer endpoints. Each “class” in the Cramer scheme is associated with a specified human exposure level, below which the risk to human health is considered as negligible. Within COSMOS the use of the traditional Cramer scheme for classifying compounds will be updated with a scheme that is not only more relevant to cosmetic ingredients but also based around modes of action, building on the mechanistic knowledge accumulated in the decades since the original publication. The initial outcome of some of these analyses has already been published by Worth et al. (2012).

The greatest part of the computational modelling of repeated dose toxicity data, specifically the identification of organ level effects and the prediction of NO(A)ELs and Lowest Observable (Adverse) Effect Levels (LO(A)ELs) will begin in COSMOS when the database is ready for use. However, there has already been much progress. Specifically, workflows are being created that enable relevant categories to be formed that allow for read-across predictions. At this time these are being developed for specific organ toxicities with an emphasis on building categories based on mechanisms of action. Within these categories there is evidence that simple read-across can be undertaken to estimate NO(A)ELs (Nelms et al. 2012). Later work in COSMOS will derive more quantitative models for predicting chronic toxicity.

The computational prediction of biokinetics is vital for successful application of in vitro to in vivo extrapolation methods. Within COSMOS physiologically-based pharmacokinetic (PBPK) models have already been developed to predict organ level concentrations for parent compound and metabolites in both rat and human. PBPK models in combination with cell based assays (that incorporate aspects of chemical fate, cell growth, toxicity and feedback) enable realistic estimates of in vivo concentration from in vitro assay data (in vitroin vivo forcing (IVIVF) techniques). The results of the first case study show this is a promising approach to extrapolating from in vitro systems to in vivo exposure. Future work will extend these models and demonstrate their applicability.

The final, and particularly significant, aspect of the COSMOS project is the development of open-access, computational workflows that can integrate access to the database and application of the models developed. The open source KNIME technology is being used for this purpose ( AG is also a partner in COSMOS). KNIME is a computational platform that enables users to create workflows and then publically distribute them so that the models can be applied by other users of the freely available KNIME software and also through a web portal without installation of the software. The workflows are flexible and expandable to the user’s requirements. KNIME also integrates many freely available computational chemistry tools (e.g. for chemical drawing, physico-chemical property calculation, statistical analysis, etc.). Thus, a workflow could allow a user to enter a compound of interest, search for that compound (or similar compounds) in COSMOS DB, run the compound through the COSMOS profilers to form a category of similar compounds (e.g. using a mechanistic profiler developed within COSMOS to identify compounds that potentially share the same mechanism of action) and perform a read-across prediction. In the same manner, biokinetics algorithms are already being built in KNIME workflows and being made freely available to anyone who installs the KNIME software. Further information on KNIME is available here.

The Future of COSMOS

COSMOS is a fast moving project that will provide computational solutions to assist the prediction of repeat dose toxicity and lay the foundations for informatics for the next generation of toxicology. Neither COSMOS, nor Seurat-1, will provide the final replacement for animals in chronic toxicity; however, a number of key computational themes are emerging from COSMOS which should inform and direct future projects.

The first key theme is integration and the pooling of information, expertise and research effort from across the world. A unified approach to informatics is enabling the creation of a single, open access database of repeat dose toxicity data with sufficient detail to act as a gold standard for all future work.

The second theme is appropriate modelling of organ level toxicity, which is informed by knowledge of biological pathways, mechanisms and modes of action and the interactions between these systems, rather than correlative approaches. A more transparent and justifiable approach to prediction will serve to increase uptake and acceptability of the models developed.

Thirdly, there has recently been increased interest the concept of the Adverse Outcome Pathway (AOP or Mode of Action) philosophy (Schultz, 2010). This provides a framework to understand and organise key events from the interaction of a chemical with a biological macromolecule to a downstream organism (or population) level response. The AOP framework provides a great cross-cutting platform demonstrating the links between all factors involved in eliciting a biological response including aspects of chemistry, physiology and biokinetics. Recently COSMOS has proposed its strategy for the adoption of AOPs (the strategy will be formally released in late 2012). Whilst AOPs were not originally foreseen as part of the COSMOS project, a cross-cutting theme within the Seurat-1 cluster to obtain mechanistic information to inform our modelling effort is envisaged. The use of AOPs as a framework to organise information from the initiating event (i.e. providing a mechanistic understanding of the chemistry involved in the initial interaction between chemical and biological system) to the final outcome is a logical step forward (Cronin et al., 2012).

Making the Pace in a Fast-Moving Field

Radical changes are underway in terms of the generation, storage and use of toxicological data and COSMOS is adaptable to these new initiatives. There is close liaison between the Seurat-1 cluster and those involved in the Tox 21 program within the US. Tox 21 is devoted to generating information to better understand toxicity pathways. Close collaboration is ensuring best use of information with integration and not duplication of effort. Work within the SEURAT-1 cluster is concentrating on organ level toxicity with an emphasis on the liver. This work is within the remit of the SEURAT-1 Mode of Action Working Group which is led by COSMOS partners. COSMOS is also playing a crucial role in leading the SEURAT-1 Biokinetics Working Group. This is also linked to the utilisation of AOPs, and more specifically prediction of the points of departure of toxicity pathways. In this way, COSMOS is not only providing valuable input into the modern age of toxicology but is also linking it to progress inside and outside of SEURAT-1.


The funding of the EU COSMOS Project by the European Community’s Seventh Framework Programme (FP7/2007-2013) under grant agreement n° 266835 and by Cosmetics Europe is gratefully acknowledged.
For more information on COSMOS see the website or refer to chapters to Chapter 4.5 in both Volumes 1 and 2 of the SEURAT-1 Annual Reports. The reports are available, free of charge, here. Enquires can be made to Dr. Andrea Richarz (

©2012 Mark T.D. Cronin, Judith C. Madden, and Andrea-Nicole Richarz

Adler, S., Basketter, D., Creton, S., Pelkonen, O., van Benthem, J., Zuang, V., et al. (2011). Alternative (non-animal) methods for cosmetics testing: Current status and future prospects – 2010. Arch. Toxicol. 85, 367-485.

Cramer, G.M., Ford, R.A. & Hall, R.L. (1978). Estimation of a toxic hazard – A decision tree approach. Food Cosmet. Toxicol. 16, 255-276.

Cronin, M.T.D. & Richarz, A.-N. (2012). Mode of Action Working Group: Use of Mode of Action related to repeated dose systemic toxicity – A framework for capturing information. In: Gocht, T. & Schwarz, M. (eds). “SEURAT-1: Towards the Replacement of In Vivo Repeated Dose Systemic Toxicity Testing. Volume 2”, pp 284-289.

Munro, I.C., Ford, R.A., Kennepohl, E. & Sprenger, J.G. (1996). Correlation of structural class with No-Observed-Effect Levels: A proposal for establishing a Threshold of Concern. Food Chem. Toxicol. 34, 829-867.

Nelms, M.D., Enoch, S.J., Fioravanzo, E., Madden, J.C., Meinl, T., Richarz, A.-N., et al. (2012). Strategies to form chemical categories from adverse outcome pathways. In: Gocht, T., Schwarz, M., (eds) “SEURAT-1: Towards the Replacement of In Vivo Repeated Dose Systemic Toxicity Testing. Volume 2” pp 263-267.

Schultz, T.W. (2010). Adverse Outcome Pathways: a way of linking chemical structure to in vivo toxicological hazards. In: Cronin, M.T.D., Madden, J.C. (eds) In Silico Toxicology: Principles and Applications. Royal Society of Chemistry, Cambridge, England. pp 346-371.

Worth, A., Cronin, M.T.D., Enoch, S., Fioravanzo, E., Fuart-Gatnik, M., Pavan, M., Yang, C. (2012). Applicability of the Threshold of Toxicological Concern (TTC) approach to cosmetics – Preliminary analysis. JRC report EUR 25162 EN. Publications Office of the European Union, Luxembourg. Available here.