AG Talbot
The "Preclinical Data Science" working group is a translational link between preclinical and clinical research. The focus here is on quantitative methods in the field of basic research. These include, for example, the development of new algorithms for processing biostatistical tasks, the application of machine learning for pattern recognition in complex biological data and the statistical planning, monitoring and evaluation of biomedical (preclinical) studies.
A particular focus of the working group is the development of methods for quantitative stress assessment of laboratory animals as well as evaluation and validation procedures in the context of multi-center studies (e.g. in the development of animal-free methods in the sense of the 3R principle).
(Dr. Steven R. Talbot, AG Head)
Preclinical data science & biostatistics
Preclinical research faces numerous challenges. It paves the way for large and important biomedical studies. High-quality basic research is a central element in gaining scientific knowledge. Particular attention is therefore paid to the scientific quality of the experimental designs, the results generated and their evaluation. Modern data science is essential for these points. As a translational link, this discipline can not only support and accompany preclinical research on many levels, but also make it clinically usable for humans with the help of quantitative methods.
The controversial topic of animal testing is a particular focus here. Researchers must decide whether and in what form the use of animals is necessary or justified for their research objectives. In line with the 3Rs principle ("reduce", "replace", "refine"), alternative methods should be explicitly considered. Where this does not make sense, pain, suffering and harm to the test animals must be minimized or avoided altogether. But how can stress be measured as objectively as possible? The working group addresses this question from a data science perspective. The question of severity assessment is not only essential for high-quality basic research, it also has enormous potential, e.g. in the automated monitoring of animals and humans in a clinical environment.
In addition, careful experimental planning is necessary to gain reliable knowledge. For example, the number of animals must be reduced to the scientifically necessary minimum. This requires careful statistical experimental design, accompanying instructions during the conduct of the study and, ultimately, support with biomedical/statistical data analysis. Only through a well-coordinated research process can reliable findings be obtained within the framework of good scientific practice (GWP). Sharpening, researching and supporting this process are key tasks of preclinical data science.
Current research
Under EU Directive 2010/63/EU, a prospective exposure assessment must be carried out in every animal testing application. Furthermore, animals must be monitored during the experiment. But how can stress be objectively monitored, measured and contextualized?
As part of the DFG-funded research group FOR2591 Severity Assessment[www.severity-assessment.de], we are developing solutions, algorithms and procedures for quantitative stress assessment.
The Severity Toolbox developed by us provides tools (R packages) that enable researchers to measure, compare and evaluate stress in animals in different ways. In particular, the Relative Severity Assessment Score (RELSA)[https://talbotsr.com/RELSA/index.html] should be mentioned here, which fuses multidimensional input signals and can be used to quantitatively compare individual animals and groups as well as animal models.
Clinical patient monitoring generates a variety of parameters over different time periods that depend not only on the type of disease, but also on several other factors such as treatment. Although the derailment of individual parameters is easy to monitor medically, the parallel evaluation of many parameters leads to a situation that is difficult to control.
In this translational project, the Patient Vital Status (PVS) developed by us takes up the idea of a fusion score and applies it to human clinical data. The aim of the project is to represent an automated monitoring of the stress state of individual patients in one value, so that multidimensional changes in state are regularized and contextualized. This enables the parallel monitoring and quantification of measurement and vital sign derailments.
The project was partially funded by the Else-Kröner Fresenius Foundation (EKFZ).
The reproducibility crisis in the medical and life sciences has shown that study results are often not sufficiently robust, which limits the scientific knowledge they provide. In particular, in vivo cell culture models, such as organoids, can exhibit large experimental variance, which can complicate both statistical design and analysis.
The aim of this project is therefore to statistically validate the experimental designs and to carry out internal and external validation procedures in the context of round robin tests and multi-center studies, both for animal models and for alternative systems to animal experiments. This is intended to clarify robustness estimators and sources of experimental variance.
The project is part of the "Micro-Replace Systems" research network and is funded by the state of Lower Saxony (zukunft.niedersachsen).
https:// r2n.eu
Teaching
We offer curricular basic courses in preclinical statistics through the Institute of Laboratory Animal Science at the MHH. If you are interested in a separate course, please contact us.
- Statistics I - Statistics and biometric methods in laboratory animal science
- Statistics II - Statistical/biometric planning of animal experiments - "Power course"
You can find the courses here.(https://www.mhh.de/tierlabor/lehr-und-sachkundeveranstaltungen/curriculare-lehre-aufbaumodule)
Publications
- Talbot, S.R., Kumstel, S., Schulz, B. et al. Robustness of a multivariate composite score when evaluating distress of animal models for gastrointestinal diseases. Sci Rep 13, 2605 (2023).
- Maria Reiber, Lara von Schumann, Verena Buchecker, Lena Boldt, Peter Gass, André Bleich, Steven Roger Talbot, Heidrun Potschka. Evidence-based comparative severity assessment in young and adult mice. PLOS ONE 18(10): e0285429 (2023).
- Maria Reiber, Lara von Schumann, Verena Buchecker, Lena Boldt, Peter Gass, André Bleich, Steven Roger Talbot, Heidrun Potschka (2023). Evidence-based comparative severity assessment in young and adult mice. PLOS ONE 18(10): e0285429.
- Tix L, Ernst L, Bungardt B, Talbot SR, Hilken G, Tolba RH (2023) Establishment of the body condition score for adult female Xenopus laevis. PLoS ONE 18(4): e0280000.
- Talbot SR, Struve B, Wassermann L, Heider M, Weegh N, Knape T, Hofmann MCJ, von Knethen A, Jirkof P, Häger C and Bleich A (2022) RELSA-A multidimensional procedure for the comparative assessment of well-being and the quantitative determination of severity in experimental procedures. Front. Vet. Sci. 9:937711.
- Alice Rovai, BoMee Chung, Qingluan Hu, Sebastian Hook, Qinggong Yuan, Tibor Kempf, Florian Schmidt, Dirk Grimm, Steven R. Talbot, Lars Steinbrück, Jasper Götting, Jens Bohne, Simon A. Krooss & Michael Ott. In vivo adenine base editing reverts C282Y and improves iron metabolism in hemochromatosis mice. Nat Commun 13, 5215 (2022).
- Schmidt T, Meller S, Talbot SR, Berk BA, Law TH, Hobbs SL, Meyerhoff N, Packer RMA and Volk HA (2022) Urinary Neurotransmitter Patterns Are Altered in Canine Epilepsy. Front. Vet. Sci. 9:893013.
- Lisa Ernst, Stefan Bruch, Marcin Kopaczka, Dorit Merhof, André Bleich, René H. Tolba & Steven R. Talbot. A model-specific simplification of the Mouse Grimace Scale based on the pain response of intraperitoneal CCl4 injections. Sci Rep 12, 10910 (2022).
- Zentrich E, Talbot SR, Bleich A and Häger C (2021) Automated Home-Cage Monitoring During Acute Experimental Colitis in Mice. Front. Neurosci. 2021; 15:760606.
- Helgers, S.O.A., Talbot, S.R., Riedesel, A. et al. Body weight algorithm predicts human endpoint in an intracranial rat glioma model. Sci Rep 10, 9020 (2020).