Image from left to right: (Top row) Anthony Lynch, Azmina Mather, Ciaran Fisher, Claire Teague, Crysthiane Ishiy, (Middle row) Daniel Sevin, Deidre Dalmas, Eric Rossman, Jan Roger, Jim Harvey, (Bottom row) Johanna Vappiani, Khuram Chaudhary, Kylie Beattie, Philippa Allen, Valeriu Damian.

How has GSK been involved in the TransQST project?
We have actively contributed to each of the organ toxicity work packages (WP5-liver, WP6-kidney, WP7-heart and WP8-gastrointestinal system). Our contribution has included providing legacy datasets, data analysis, modelling, and the generation of novel data (using established and novel technologies) where the need has been identified. All this, as well as input from GSK subject matter experts, has contributed to furthering the development and validation of quantitative systems toxicology (QST) models within the project and driving the direction of the modelling strategy.

As an industry partner, engaged across the whole project, one of our key contributions has been to provide an industry perspective of the constraints, challenges, and core requirements for the successful application of systems modelling within a pharmaceutical industry environment for the assessment of drug safety. Through actively contributing our perspective and expertise, we’ve aimed to help guide the project to ensure the outputs of the TransQST consortium drive the state of the art forward, are relevant to the needs of the industry, fit-for-purpose, and ultimately have the potential to enhance the delivery of safe drugs for the benefit of patients.

What are GSK’s main achievements in TransQST?
GSK led the incorporation of a number of novel techniques into the project. GSK bio-imaging experts applied spatial transcriptomics methodologies to image the rat kidney and quantify localised responses across the tissue.  We also commissioned the generation of novel data using the NewCells in vitro aProximate™ platform and the application of this data for informing modelling is currently being explored. GSK scientists have co-authored a number of publications and conference presentations arising from TransQST activities, with several more publications in preparation or planned. We have also shared a significant amount of legacy data to support the modelling efforts in the project, through the data storage tools developed by partner EMBL-EBI to make this available to TransQST consortium.

What are the main challenges that the GSK team has faced across the project?
As is common with ambitious project proposals and large multi-disciplinary teams, it took time to define what could be achieved realistically within the project. Adding to this, at the start of the project there was a sparsity of relevant data for modelling in some of the target organs. It took time for the GSK team to identify appropriate data sets which could be contributed or to generate novel data to support the efforts of the TransQST consortium. Some of the compounds selected as exemplars to explore within the project presented some additional technical challenges and resulted in further obstacles to overcome.

From the start of the project there was also some level of disconnect between the reality of the data produced in standard pharmaceutical industry studies and the types of data required (and believed that would be readily available) to derive model parameters and inputs. QST models typically require more detailed timepoints and additional endpoints than are collected in our routine experimental studies. Helping to resolve this disparity in perception has been a key contribution of the GSK and other industry partners. This is an on-going challenge that continues to need to be addressed to allow the full demonstration and application of TransQST tools and models to compound examples in active pharmaceutical development which, if achieved, would be an ideal conclusion to the project. GSK has worked together with the consortium to make progress towards addressing these challenges.

In the development and testing of which tool/models has your institution been involved in and what is their impact?
We have been directly involved in modelling and data analysis in the gastrointestinal and renal work packages as well as providing direction and data contributions for modelling efforts in the cardiac and liver work packages. Within GSK we have also tested a number of tools arising from the TransQST project including the WGCNA tool, the R Shiny hemodynamic modelling tool (the University of Leiden are leading the development of both of these tools), and the Virtual Assay cardiac modelling platform (which has been developed by the University of Oxford). We are also in the process of further evaluating the TransQST cardiac contractility model with GSK compound data. These models and tools, and our involvement in the development of new modelling approaches, has allowed us to more clearly demonstrate within GSK how non-clinical data can be used as inputs to modelling, how models can be used to integrate various different data source and be applied to predict toxicity liability in different organ systems and situations. By being able to showcase the potential of modelling in this way we have also been able to expand our modelling resource so that we can make this identified potential a reality. Our involvement in TransQST has also reinforced the utility of in vitro data for modelling, which itself has led to further investment in in vitro model development to support modelling efforts. 

Where specifically does GSK see the outputs of TransQST being able to impact pharmaceutical drug development?
We want to continue to embed modelling as a routine part of non-clinical safety assessment, using models to integrate in vitro, in vivo and in silico data to predict safety concerns earlier, elucidate mechanisms of toxicity and ultimately reduce attrition. We want models to be used to add a weight of evidence and provide quantitative predictions to accompany early safety flags and ultimately to improve our confidence in therapeutic index and our ability to translate observations in non-clinical species to clinical liability for toxicity. Our involvement in TransQST increased our belief that this ambition is possible and enabled us to move a few steps further on that journey. 

What do you believe is the main contribution of TransQST in the field?
The main contribution from our perspective is bringing together the community of industry, regulatory and academic partners all with an interest in predictive modelling of toxicity liability including physiologists, clinicians and modellers to allow for a truly multi-disciplinary approach to QST model development. Connections and collaborations have been made through this consortium which we expect will be maintained beyond the project. Involvement in the TransQST project has allowed the ‘state of the art’ for quantitative systems toxicology to be better defined and formed a foundation from which to move that state of the art forward. 

If the project was to start again, how might we approach the project given the learnings we’ve gained from being involved in TransQST?
We would begin the project with a thorough review of the current state of the art for quantitative systems toxicology, and identify tractable organ toxicities where there are high industry attrition rates, but historically less application of modelling (e.g modelling of the central nervous system).  We would also aim for better alignment from the project outset between modelling inputs which would be essential and endpoints which are routinely generated in pharmaceutical studies. In this way we would hopefully have a more robust set of data readily available to support model development and validation and we would be more easily able to test the models in prospective use cases in industry. Similarly, a critical evaluation of exemplar compounds used for model development would be undertaken, to try to select those which would be both informative for model development and technically tractable. Having the experience of the end of the project, where we are navigating license agreements for TransQST tools we wish to use internally, we would also use what we have learnt through this process to try to make this path smoother from the outset.