Walter S. Woltosz

Predicting the impact of physiological and biochemical processes on oral drug bioavailability.

Recent advances in computational methods applied to the fields of drug delivery and biopharmaceutics will be reviewed with a focus on prediction of the impact of physiological and biochemical factors on simulation of gastrointestinal absorption and bioavailability. Our application of a gastrointestinal simulation for the prediction of oral drug absorption and bioavailability will be described. First, we collected literature data or we estimated biopharmaceutical properties by application of statistical methods to a set of 2D and 3D molecular descriptors. Second, we integrated the differential equations for an advanced compartmental absorption and transit (ACAT) model in order to determine the rate, extent, and approximate gastrointestinal location of drug liberation (for controlled release), dissolution, passive and carrier-mediated absorption, and saturable metabolism and efflux. We predict fraction absorbed, bioavailability, and C(p) vs. time profiles for common drugs and compare those estimates to literature data. We illustrate the simulated impact of physiological and biochemical processes on oral drug bioavailability.

Simulations of the Nonlinear Dose Dependence for Substrates of Influx and Efflux Transporters in the Human Intestine

The purpose of this study was to develop simulation and modeling methods for the evaluation of pharmacokinetics when intestinal influx and efflux transporters are involved in gastrointestinal absorption. The advanced compartmental absorption and transit (ACAT) model as part of the computer program GastroPlus™ was used to simulate the absorption and pharmacokinetics of valacyclovir, gabapentin, and talinolol. Each of these drugs is a substrate for an influx or efflux transporter and all show nonlinear dose dependence within the normal therapeutic range. These simulations incorporated the experimentally derived gastrointestinal distributions of transporter expression levels for oligopeptide transporters PepT1 and HPT1 (valacyclovir); System L-amino acid transporter LAT2 and organic cation transporter OCTN1 (gabapentin); and organic anion transporter (OATP1A2) and P-glycoprotein (talinolol). By assuming a uniform distribution of oligopeptide transporter and by application of the in vitro Km value for valacyclovir, the simulations accurately reproduced the experimental nonlinear dose dependence. For gabapentin, LAT2 distribution produced simulation results that were much more accurate than OCTN1 distributions. For talinolol, an influx transporter distribution for OATP1A2 and the efflux transporter P-glycoprotein distributed with increasing expression in the distal small intestine produced the best results. The physiological characteristics of the small and large intestines used in the ACAT model were able to accurately account for the positional and temporal changes in concentration and carrier-mediated transport of the three drugs included in this study. The ACAT model reproduced the nonlinear dose dependence for each of these drugs.

Prediction of Modified Release Pharmacokinetics and Pharmacodynamics from In Vitro, Immediate Release, and Intravenous Data

The aim of this study was to demonstrate the value of mechanistic simulations in gaining insight into the behaviors of modified release (MR) formulations in vivo and to use the properly calibrated models for prediction of pharmacokinetics (PK) and pharmacodynamics (PD). GastroPlusTM (Simulations Plus, Inc.) was used to fit mechanistic models for adinazolam and metoprolol that describe the absorption, PK, and PD after intravenous (i.v.) and immediate release (IR) oral (p.o.) administration. The fitted model for adinazolam was then used to predict the PD profile for a MR formulation and to design a new formulation with desired onset and duration of action. The fitted metoprolol model was used to gain insight and to explain the in vivo behaviors of MR formulations. For each drug, a single absorption/PK model was fitted that provided simulated plasma concentration–time profiles closely matching observed in vivo profiles across several different i.v. and p.o doses. Sedation score profiles of adinazolam were fitted with an indirect PD model. For metoprolol, the fitted absorption/PK model for IR p.o. doses was used to select in vitro dissolution conditions that best matched the in vivo release of MR doses. This model also explained differences in exposure after administration of MR formulations with different release rates. Mechanistic absorption/PK models allow for detailed descriptions of all processes affecting the two drugs’ bioavailability, including release/dissolution, absorption, and intestinal and hepatic first pass extraction. The insights gained can be used to design formulations that more effectively overcome identified problems.

Modeling of Absorption

Absorption takes place when a compound enters an organism, which occurs as soon as the molecules enter the first cellular bilayer(s) in the tissue(s) to which is it exposed. At that point, the compound is no longer part of the environment (which includes the alimentary canal for oral exposure), but has become part of the organism. If absorption is prevented or limited, then toxicological effects are also prevented or limited. Thus, modeling absorption is the first step in simulating/predicting potential toxicological effects. Simulation software used to model absorption of compounds of various types has advanced considerably over the past 15 years. There can be strong interactions between absorption and pharmacokinetics (PK), requiring state-of-the-art simulation computer programs that combine absorption with either compartmental pharmacokinetics (PK) or physiologically based pharmacokinetics (PBPK). Pharmacodynamic (PD) models for therapeutic and adverse effects are also often linked to the absorption and PK simulations, providing PK/PD or PBPK/PD capabilities in a single package. These programs simulate the interactions among a variety of factors including the physicochemical properties of the molecule of interest, the physiologies of the organisms, and in some cases, environmental factors, to produce estimates of the time course of absorption and disposition of both toxic and nontoxic substances, as well as their pharmacodynamic effects.

Novel ADMET design tool for chemists

We present a new tool for molecule design called ADMET Sketcher™. The tool allows chemists to draw any number of molecules inside of a canvas, and immediately obtain predicted values of dozens of key ADMET properties from our best-in-class ADMET Predictor™ software package. The sketcher itself includes a number of novel capabilities, including a feature to modify both the bond angles and torsional angles of molecule side chains, and an advanced structure-cleanup feature with the option to keep one or more regions of the molecule fixed in place. Yet the true power of the tool comes from the numerous ADMET predictions covering physicochemical, biopharmaceutical, metabolism, and toxicities, which are updated dynamically as structures are edited, allowing near instantaneous feedback about which structural changes affect which properties, and in what manner. The ability to handle multiple structures means that a molecule found to have a desirable property profile can be kept as a reference, so properties of subsequently drawn molecules can quickly and easily be compared with the original. This is particularly useful when designing analogs of a known lead. We illustrate the tool by examining a diverse set of structural changes within a collection of molecules active against the HIV Integrase enzyme.

Compressor Performance Modeling and Prognostics Using Artificial Neural Networks

Data collected routinely from engine operation is used to construct a compressor performance map through regression modeling using Artificial Neural Network Ensembles (ANNEs). Such a map can be updated frequently and accessed in real time to detect shifts in the compressor performance. Such shifts can be indicators of degradation, maintenance issues, and impending faults. Updating the map can provide dynamic updates to the operating limits for compressor operation in the engine controller to steer clear of the surge line. Additionally, the resulting model could be used to estimate remaining useful life.

Near real-time characterization of unknown missiles in flight using computational intelligence

This paper focuses on the rapid characterization and identification of missiles of both known and unknown types early in their trajectories. Many physical relationships in the field of aerospace and aeronautics can be used to create computational models to calculate characteristics of an aerodynamic system. In such cases, the models created are generally more computationally intensive for problems of higher complexity. For example, computational fluid dynamics (CFD) yields high-fidelity solutions at the expense of long computation times. In a missile defense scenario where seconds are critical, computations must be performed much more rapidly than CFD while maintaining a high level of accuracy. In the event that an adversarial missile system of an unknown type is launched, there is an even greater need for rapid characterization. Computational intelligence methods provide a means to determine the underlying relationships between sets of data that may not be obvious to a human observer, such as between missile kinematic data and missile geometric data. Once the computational cost of training the algorithms has been invested, the calculation time per new solution is reduced to the order of milliseconds, enabling near real-time applications. In this study, we adapted a mature artificial neural network ensemble (ANNE) methodology originally developed for pharmaceutical research to accurately predict the diameter of missiles based on critical points of their ballistic trajectories, such as the state of the missile at motor burnout. The results indicate that it is feasible to determine geometric parameters, such as missile diameter, based on sparse telemetry data while a missile is still in flight. We expect that additional independent models can be trained for fineness ratio and other geometric measures. We also expect that the difficulties that early boost termination and other forms of “sandbagging” present for diameter prediction can be overcome. With further enhancement, this methodology could serve as an additional input to current missile defense algorithms, and is particularly well-suited as a supplementary process for characterizing previously unknown missiles.

Rapid calculation of missile aerodynamic coefficients using artificial neural networks

A variety of machine-learning methods has been applied to problems for which physics-based solutions are either nonexistent or computationally expensive. Based on such methods, surrogate models, i.e., empirical models that are trained on outputs of the more computationally intensive methods, can provide acceptable accuracy while dramatically reducing execution time and expense. This work describes the application of an artificial neural network ensemble (ANNE) approach, originally developed over the past fifteen years for cheminformatics studies in the pharmaceutical industry, to train surrogate models that predict missile aerodynamic coefficients. This modeling engine has been consistently demonstrated to provide best-in-class predictive models for a variety of molecular properties from only a molecules structure, and is in widespread use in the pharmaceutical industry today. The surrogate models developed to predict aerodynamic coefficients for arbitrarily shaped missiles at arbitrary Mach numbers and angles of attack have resulted in highly accurate predictions that execute in milliseconds on a modern laptop computer. The ability for rapid predictions can be integral in the design process for missiles and other aerodynamic bodies. Building on previous work, we show how descriptor sensitivity analysis identifies the key descriptors driving the model performance, and relates inputs to outputs to help meet critical design/mission objectives.