A new approach to addressing two problems in pharmacokinetics and pharmacodynamics using machine learning
Date
2020
Authors
Habib, Sohaib, author
Reisfeld, Brad, advisor
Munsky, Brian, committee member
Shipman, Patrick, committee member
Journal Title
Journal ISSN
Volume Title
Abstract
In this work, machine learning was applied to develop solutions for two problems related to drug pharmacokinetics (PK) and pharmacodynamics (PD). The first problem was finding a way to easily predict important pharmacological measures accurately representative of those from simulation results computed via a sophisticated model for drug absorption via oral dosing. This model (OpenCAT: Open source Compartmental And Transit model) comprises a system of differential equations describing the absorption of drugs into the gastrointestinal tract, including such factors as drug dissolution and spatially-distributed absorption, metabolism, and transport. For this problem, a machine learning framework was built to develop a self-contained random forest representation of the model predictions that could be queried for critical PK parameters such as maximum plasma concentration (Cmax), time at which the maximum concentration occurs (tmax), and the area under the concentration-time curve (AUC). The random-forest representation was able to generate predictions for the targeted PK parameters close to the solution of the original OpenCAT model over a wide range of drug characteristics. The second problem involved predicting the pharmacodynamics (cholinesterase reactivation) of antidotes for nerve agents. In this case, a machine learning framework was built to use experimental data and corresponding theoretically-derived chemical descriptors to predict the pharmacodynamics of new candidate antidotes against both tested and untested nerve agents. Overall, this project has demonstrated the utility of machine learning approaches in the fields of drug pharmacokinetics and pharmacodynamics.
Description
Rights Access
Subject
nerve agent
pharmacology
XGboost
PBPK models
machine learning
random forest