Bioinformatics-inspired non-parametric modelling of pharmacokinetics-pharmacodynamics systems using differential neural networks

Bionformatics and pharmacokinetics-pharmacodynamics (PKPD) systems are two conjugated tools to intensively explore the effect of new drugs on the human body running in-silico analysis. Usually, PKPD models do not consider all the biological reactions that explain the pharmaceutical effect. A complementary non-parametric modeling can be useful to recover the PKPD dynamics despite the uncertainties and external perturbations effect, which can reduce the degree of uncertainties on the drug evaluation. The aim of this study is to get a feasible non-parametric model of PKPD models using a bioinformatics inspired evaluation of antibacterial drug doses. A class of bioinformatics inspired differential neural networks (DNNs) responding to the dose modification provides the non-parametric approximation of the PKPD dynamics. The DNN modeling strategy was applied to approximate the dynamics of PKPD models under four different dosing regimes. The modeling strategy estimated the bacteria survival (measured as the logarithm of the colony forming units per milliliter) after the drug application. The same adjusted DNN-based model confirmed the ability of designing an off-line lab for evaluating diverse dosing strategies of antibacterial pharmaceutical.