Translation of pharmacometric models from NONMEM to nlmixr2 and RxODE2

Sammanfattning: The gold standard for pharmacometrics modeling, along with its modeling format, is currently NONMEM. In order to use other software, there is often a manual step of converting a model from one format to another, which is both time-consuming and causes manual errors. This project aimed to solve this problem by creating a conversion- and validation tool from NONMEM to two formats: nlmixr2 and RxODE2. These are both, unlike NONMEM, freely available and integrated into R. This was done by integrating the two tools (conversion and validation) into the program Pharmpy, which can extract model information from NONMEM's model format. For conversion, the model was read into Pharmpy and then, part by part, converted to the respective model format. The associated validation compared the predictions of the respective programs to see if they differed significantly from each other. The project showed that this type of conversion is possible, but the programs showed a greater difference than expected. Part of this can be explained by a rounding of parameter values in Pharmpy, but further analysis also indicated fundamental differences in the determination of the predictions between the programs. Larger differences in predictions are for instance oftentimes equidistant from the actual observation, meaning the predictions are presumably calculated differently. While not disproving the converted model, smaller discrepancies between the programs would indicate a more confident validation. In summary, the developed tools are considered useful for models where Pharmpy supports parsing of the model. If not for complete conversion, then at least for partial conversion with manual correction, which is also an improvement over an entirely manual workflow.