Translational quantitative systems toxicology to improve the understanding of the safety of medicines


When developing a new medicine, determining whether or not it will be safe for patients to take remains a challenge. It’s estimated that 197 000 deaths per year in the EU are caused by adverse drug reactions (ADRs) and that 5% of all hospital admissions in the EU are due to a ADR.

Some ADRs are only identified after a medicine has already been authorised. There is no ideal way to test drugs – the current drug development pipeline is a best practice, but it is not error-free.

One problem is that drugs are initially tested on animals which can react differently to humans. In fact, it’s known that animal tests fail to predict adverse reactions in humans in approximately 30% of cases. In vitro tests, although they are more human-specific, are also not ideal because these often show the effects of a drug on one particular organ whereas the human body is an entire system. For instance, one drug might be used to treat a cardiovascular problem and it is shown to have no negative effect on the heart tissue in vitro. However, in clinical trials it becomes clear that the drug has an adverse effect on the liver.

Drugs with adverse drug reactions still make it to the market however, often because of a costs/benefits assessment – for instance, many chemotherapy drugs have awful side effects but the alternative for patients is death, so the benefits outweigh the disadvantages. On the other hand, patients will tolerate very few side effects in a medication for a relatively minor illness, like a cold.

Combining in vivo, in vitro and in silico methods for the best model

Although the perfect model does not exist, the TransQST team decided to take an approach combining in vitro, in vivo and in silico techniques to develop a system of models that can better predict adverse reactions in patients.

Known as ‘quantitative systems toxicology’, the approach involves developing a computer simulation that uses the results of in vivo and in vitro modelling in an algorithm to more accurately predict and eliminate drug candidates with severe side effects. The new models developed by TransQST should be sufficiently complex to accurately mimic the human body’s response.

The project set out to develop models focused on the liver, kidney, cardiovascular and gastrointestinal-immune systems because these are the body parts that frequently experience adverse reactions. Each system had a different starting point: research on cardiovascular in vitro and in silico ADR models started in the 1970s, so the TransQST consortium had a solid amount of data to start from. For the liver and kidney, some research had been undertaken but not as much was known about these organs compared to the heart. And for the gastrointestinal-immune system, the project had to start from scratch – at the data collection stage. Despite the lack of data, gastrointestinal-immune system side effects are amongst the most common, so the TRANSQST project established themselves as pioneers in this field in particular.

At the heart of the problem

Out of all of the systems covered by TransQST, the results from the work on the heart are the closest to being used within the drug development pipeline. The in silico model produced by the project was able to predict rhythmic abnormalities in the heart caused by exposure to a drug with 90% accuracy, a significant improvement compared to the current state-of-the-art animal-based practices, which report a success rate of 75-80%. The researchers are continuing to work on validation studies to confirm the credibility of this technique, which is important for eventual regulatory approval.

Mapping impact on the liver and kidneys

The TXG-MAPr tool was developed by the project to predict how rat livers and kidneys might react to various compounds. To create the tool, the project used RNA transcripts of rat livers and kidneys and examined them using a technique called weighted correlation gene co-expression network analysis, which groups genes with similar expressions into clusters called modules. The tool uses colours to indicate how active a particular gene becomes when exposed to a compound, with red for high activity and blue or green for low activity.  The tool can also be used to examine signalling pathways which can help scientists to understand not only that toxicity is present, but also how that toxicity travels through the body. This is important because many drugs work by blocking signalling pathways within the body.

Some of the core issues with drug development pipelines are the physiological and biological differences between rats and humans. For instance, the number of nephrons (filtering units found in the kidneys) between humans and rats is not the same. The project used this information and genetic data about the species-specific differences between people and rats to develop an algorithm that translates results from rat studies on the kidneys into predictions for human reactions. This model might be particularly useful for dose selection in clinical trials.

Pioneers in models for adverse gut reactions to drugs

Data regarding the mechanisms for adverse drug reactions in the gastrointestinal system were scarce at the outset of the project, but by the time TransQST came to a close they had successfully developed several initial models predicting how drugs might damage the gastrointestinal system. Irritation of the villi, tiny cells that line the small intestine, is linked to diarrhoea, so the project developed three models evaluated the risk of diarrhoea from exposure to various drugs: one computer model, one mouse-based model and one in vitro model. These gastrointestinal toxicology models have been used in early clinical trial stages to support the development of new cancer treatments.

Reproducibility of existing all-systems mathematical models

At present, the pharmaceutical industry relies on mathematical models to predict how much of a drug is needed to have the desired clinical effect in a patient. Part of the TransQST project also looked at existing all-systems mathematical models to evaluate how robust and reproducible they were. The reproducibility of 455 models from peer-reviewed research articles in 152 journals published from 1980 to 2020 was assessed, and the finding was that only half could be reproduced. Whereas a lack of reproducibility is highly likely in other fields of science, where different equipment and laboratory processes can make reproducing another laboratory’s results difficult, mathematical models are built on equations and so should be easy to reproduce. The main problems were missing parameters, missing details about initial conditions and model structural inconsistency. However, for seventy models the reason for non-reproducibility was unclear.

Based on these studies, the TransQST project produced an 8-point reproducibility scorecard, so that authors, editors and reviewers of models can easily and quickly check whether they have contained enough information in their manuscript to make reproducing their mathematical model possible. It’s hoped that this scorecard could improve the reproducibility of all-systems mathematical models going forwards.

Reducing animal research

The models put forward by TransQST could potentially reduce the amount of animal lives that are lost during medical research. If TransQST’s models eliminate a potential drug candidate because of adverse drug reactions, a pharmaceutical company may decide that there is no point in embarking on a clinical trial for that candidate, and the amount of animals that would have died or suffered in the initial rounds of that trial will be spared.

Applications of TRANSQST models

One concrete example of how the TransQST models can speed up the drug development process arose during the COVID-19 pandemic. At the time, an existing drug hydroxychloroquine was being suggested as a potential treatment for the coronavirus. TransQST’s cardiac model indicated that a high dosage of the drug for tackling COVID-19 would have adverse effects on the heart for some patients. The model was able to identify which patient populations would be at risk for those adverse effects.

The results of the TransQST project mean that molecules that are being investigated as potential treatments can be eliminated before the clinical trial stage if the TransQST models indicate that the risk of adverse side effects is high. This will reduce costs for pharmaceutical companies and increase efficiency in the drug development pipeline.

The TransQST project published 85 peer-reviewed papers, had 90 oral or poster presentations at scientific conferences. 23 webinars were held explaining how to use various TransQST models, and the consortium was invited to showcase their results at the EU Research and Innovation Days’ Science is Wonderful! exhibition in September 2020.

In the future, researchers could build on the results from the TransQST project to understand QST modelling for direct patient benefit, in a personalised approach to clinical dosage.

Achievements & News

IMI safety project TransQST showcases activities at the R and I days

IMI’s safety project TransQST was one of the EU-funded projects featured at the virtual exhibition of the recent EU Research and Innovation Days. ###When developing a new medicine, scientists need to know if the compound will be harmful to vital organs such as the heart, liver, or kidneys. However, studying this is far from easy. TransQST is gathering data and developing computer-based tools to make it easier to assess the safety profile of potential medicines. This will both improve medicines safety and reduce the use of animals in research.

In a presentation at the event, project participant Elisa Passini of the University of Oxford explains that computer models can deliver more accurate results than animal experiments (89% for computer models compared to 75-85% for animal tests). In addition to models based on single heart cells, Dr Passini is also working on three-dimensional computer models of the heart that require a super computer to run.

In total, over 35 000 people attended the R&I days, which ran online from 22 to 24 September.

Find out more


  Show participants on map
EFPIA companies
  • Abbvie Deutschland GMBH & Co Kg, Wiesbaden, Germany
  • Astrazeneca AB, Södertälje, Sweden
  • Boehringer Ingelheim Internationalgmbh, Ingelheim, Germany
  • Eli Lilly And Company LTD, Basingstoke, United Kingdom
  • Glaxosmithkline Research And Development LTD., Brentford, Middlesex, United Kingdom
  • Institut De Recherches Internationales Servier Iris, Suresnes, France
  • Janssen Pharmaceutica Nv, Beerse, Belgium
  • Orion Oyj, Espoo, Finland
  • Sanofi-Aventis Recherche & Developpement, Chilly Mazarin, France
Universities, research organisations, public bodies, non-profit groups
  • Erasmus Universitair Medisch Centrum Rotterdam, Rotterdam, Netherlands
  • European Molecular Biology Laboratory, Heidelberg, Germany
  • Forschungsgesellschaft Fur Arbeitsphysiologie Und Arbeitsschutz Ev, Dortmund, Germany
  • Fundacio Institut Hospital Del Mar D Investigacions Mediques, Barcelona, Spain
  • The University Of Liverpool, Liverpool, United Kingdom
  • Universitat Wien, Vienna, Austria
  • Universitatsklinikum Heidelberg, Heidelberg, Germany
  • Universiteit Leiden, Leiden, Netherlands
  • Universiteit Maastricht, Maastricht, Netherlands
  • University of Oxford, Oxford, United Kingdom
Small and medium-sized enterprises (SMEs) and mid-sized companies (<€500 m turnover)
  • Certara Uk Limited, London, United Kingdom
  • Synapse Research Management Partners SL, Barcelona, Spain
Non EFPIA companies
  • Crown Bioscience Netherlands BV, Utrecht, Netherlands
  • Vertex Pharmaceuticals (Europe) Limited, London, United Kingdom

NameEU funding in €
Certara Uk Limited484 600
Crown Bioscience Netherlands BV325 354
Erasmus Universitair Medisch Centrum Rotterdam220 500
European Molecular Biology Laboratory1 093 089
Forschungsgesellschaft Fur Arbeitsphysiologie Und Arbeitsschutz Ev350 250
Fundacio Institut Hospital Del Mar D Investigacions Mediques900 525
Synapse Research Management Partners SL353 669
The University Of Liverpool1 161 914
Universitaetsklinikum Aachen (left the project)132 435
Universitat Wien350 250
Universitatsklinikum Heidelberg269 790
Universiteit Leiden1 239 324
Universiteit Maastricht793 300
University of Oxford325 000
Total Cost8 000 000