Re-design and improvement of animal experiments, using Bayesian methods

Abstract: The pharmaceutical industry uses animal models in a variety of settings including safety, pharmacodynamic modelling and efficacy. Conventionally, a frequentist design of standard animal experiments of human asthma replication includes repeatedly running treatment groups with the exact layout, focusing on effect size rather than capturing historical data of animal responses on compounds of interest. Here, we propose a Bayesian framework which is used as an alternative to the frequentist approach. This allows for the incorporation of prior beliefs into the experimental process. Specifically, non-informative, semi-informative and informative prior distributions are assigned to Single-Level Normal models. Given the priors aforementioned, it was found that using semi-informative priors leads to the creation of consistent historical data. Given these beliefs, we combine the results of all experiments by implementing a Two-Level Bayesian Meta-Analysis, achieved by adding the extra level of experimental studies to the data structure. The end point of this project showed that simulation trials of all experimental studies with the assignment of semi-informative priors can result in the reduction of animals per treatment group by a margin of 10 %.