Quantitative Systems Pharmacology and Empirical Models: Friends or Foes?

Mechanistic Models as Framework for Understanding Biomarker Disposition: Prediction of Creatinine‐Drug Interactions

“…The mechanistic dynamic models allowed quantitative integration of creatinine renal disposition, interrogation of mechanistic assumptions, and identification of knowledge gaps and uncertainties (in fraction transported, permeability data, and tubular re‐absorption). All above is consistent with quantitative systems pharmacology approach (i.e., a useful model is one that permits new mechanistic insight to be gained) 36 . In addition, dynamic models allowed simulation of time course of changes in serum creatinine, together with newly proposed prediction limits that accounted for intra‐individual variability in serum creatinine for the evaluation of prediction success of biomarker interactions.…”

Application of Machine Learning for Tumor Growth Inhibition – Overall Survival Modeling Platform

“…Another commonly cited drawback of ML approaches is generalizability, because ML models cannot extrapolate beyond feature space of the training data, whereas the traditional empirical models used in pharmacometrics and quantitative systems pharmacology in particular can be extrapolated to a certain extent. 39 Therefore, the dataset used to train ML models must be relevant to the population being studied. Due to the relatively small dimensionality of the analysis dataset by ML standards, cross validation using bootstrap 40 was performed in the current analysis instead of using k-fold cross validation, however, our investigation was designed to be exploratory, and an external validation with data from additional studies should be performed in the future to confirm the validity of the conclusions.…”

“…Hence, additional work is needed to compare the predictive performance of the ML models and the important predictors identified by the ML approaches in larger datasets, possibly by combining data from multiple clinical trials or leveraging alternative data sources, such as longitudinal tumor dynamic data instead of TGI metrics or real‐world data, as well as incorporating other covariates, although a physiological understanding of the correlation between covariates and the survival outcome should still remain, as the lack of interpretability has been one of the major criticisms of using ML approaches. Another commonly cited drawback of ML approaches is generalizability, because ML models cannot extrapolate beyond feature space of the training data, whereas the traditional empirical models used in pharmacometrics and quantitative systems pharmacology in particular can be extrapolated to a certain extent 39 . Therefore, the dataset used to train ML models must be relevant to the population being studied.…”

Sequential model-based A- and V-optimal design of experiments for building fundamental models of pharmaceutical production processes

“…Especially in pharmaceutical production processes, mathematical models offer crucial information at multiple stages in the drug development and scale-up the process, thereby saving time and effort. Empirical models are commonly used to aid the development of pharmaceutical production processes by providing equations that link experimental inputs to the outputs of interest (Benson, 2018;Boukouvala et al, 2011;Domagalski et al, 2015;Hallow et al, 2010;Huang et al, 2009). The main disadvantage of the empirical models is that they are usually not suitable for obtaining predictions for dynamic process behavior nor for predicting behavior outside the range of experimental conditions used in model development.…”