Computer Simulations and Modeling
The aim of science is, on the one hand, as complete a comprehension as possible of the connection between perceptible experiences in their totality, and, on the other hand, the achievement of this aim by employing a minimum of primary concepts and relations.
–Albert Einstein
What’s obvious isn’t always what’s true.
–John Dunning, American author.
Few areas of drug development hold out as much potential for great strides forward as does the area of computer modeling and simulations. This chapter is in the section of “Overview of Current and Future Development” for that very reason. Progress in this area is going forward, and many proponents are extolling the virtues of using these approaches in many areas of research and development (R and D). On the other hand, there are skeptics who say that however good your models, they are only models and cannot accurately include human behavior into the model or simulation. Their view is that trying to predict human behavior is as difficult today as it was 100 years ago. Another difficulty in creating accurate simulations is that regulatory requirements may not be able to be incorporated into the earlystage models prior to meetings with regulators and learning their reactions, which are notoriously hard to predict. However, this chapter shows the potential that these technologies have and the areas in R and D in which models and simulations are being developed and used.
DEFINITIONS AND APPROACHES TO COMPUTER SIMULATIONS AND MODELING
Definitions
The term modeling implies a mathematical approach or construct built from basic processes or data relating inputs to outputs that is made to predict something of value for a drug. The “something” may be a prediction of absorption, distribution, metabolism, or excretion (i.e., a pharmacokinetic parameter), or it may be a pharmacodynamic aspect of how the compound or drug may act in a pharmacological study or in human clinical trials. Models are created with existing data and, therefore, are looking backwards in time.
The term simulation refers to what is built with mathematical models, as described earlier, and then used so that one or more variables of interest may be randomly varied to determine
the output or result of interest. Therefore, a simulation looks forward in time. An entire clinical trial could be simulated to see what the primary outcome would be if one controlled (by simulation) all other characteristics of the trial, possibly including patient responses. Table 114.1 illustrates the similarities and differences between modeling and simulations.
the output or result of interest. Therefore, a simulation looks forward in time. An entire clinical trial could be simulated to see what the primary outcome would be if one controlled (by simulation) all other characteristics of the trial, possibly including patient responses. Table 114.1 illustrates the similarities and differences between modeling and simulations.
Table 114.1 Similarities and differences between modeling and simulation | |||||||||||||||||||||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| |||||||||||||||||||||||||||||||||||
An alternative way of conducting a simulation is to choose a number of different inputs that are not randomly chosen but are specific numbers that are believed (or possibly known) to be realistic for the system being simulated. These may be in the field of physiology or economics where different scenarios are identified.
The term bioinformatics was defined by Kantardjieff and Rupp (2004) as a “complex landscape of computational methods that utilize data mining, machine learning techniques, and predictive modeling to provide discovery of information or knowledge useful” to develop new drugs. They go on to say that bioinformatics obtains sequence-derived information from genomic databases and detailed three-dimensional structure information from crystallographic, nuclear magnetic resonance or homology modeling methods and in silico screening of molecular libraries. Thus, it examines the flow of information “from genome to structure to function.”
Stochastic versus Deterministic Simulations
One may posit that there are two types of simulations from a statistical or methodological perspective: those that are stochastic, and those that are deterministic. Stochastic simulations involve using random variables, which thus involves chance or probability in what is chosen to create the model or simulation. A “stochastic variable” is defined as a random variable, and a “stochastic process” is one that involves chance or probabilistic events or processes. This approach is used a great deal by statisticians. For example, they will often generate random numbers by postulating distributions for the inputs when creating a simulation. Then they obtain a variety of outputs that are then evaluated.
Deterministic simulation denotes that a range of specific values is chosen for use in the simulation, and the outputs are referred to as scenarios. The values are chosen systematically to represent the most likely range of values that would be obtained in the specific situation. The outputs or scenarios are then evaluated on the basis that “if X occurs (i.e., the input), then scenario A (i.e., the output) will be obtained, and if Y occurs, then scenario B will be obtained.”
Overview of Computational Modeling in Research and Development
Kumar et al. (2006) have illustrated the way that computational modeling fits with traditional biology and informatics in the R and D areas of a pharmaceutical company. Their figure (Fig. 114.1) shows several relationships that illustrate an approach to hypothesis generation and predictions. While this will not be everyone’s approach to viewing the interrelationships among these groups and functions, this is one approach that can be used as an introduction to understanding the area of developing models. A related overview figure was presented by Kitano (2002). This use of simulation for training professionals is a situation where the deterministic approach would be preferable to use versus the stochastic approach.
Top-down versus Bottom-up Approaches to Physiological Modeling
The main strategy to develop models of physiological and eventually clinical responses to drugs and to seek to understand and predict responses is to consider approaching this by a top-down or a bottom-up approach. These approaches are reviewed by Michelson, Sehgal, and Friedrich (2006).
In the top-down approach, one initiates one’s thinking from the perspective of a clinical event or observation and whatever understanding one has of the overall organism. Michelson, Sehgal, and Friedrich (2006) continue: “For any given disease process, once clinically relevant behaviors have been identified, the modeling process begins by identifying those subsystems required to reproduce them. The physiological and biological components that make up those subsystems are then modeled. This process continues in an iterative fashion, adding greater detail to each subsystem of the model (Friedrich and Paterson 2004). In this way, each subsystem is constrained by the overall behavior of the entire system.” Michelson (2003) mentions that the top-down approach allows researchers to pose “what if” scenarios and to identify gaps and inconsistencies in the model, which leads to improved data collection.
Entelos is a company based in Foster City, California, that has published several papers on this approach and how it works with clients to simulate aspects of a clinical trial (Bangs 2005; Kansal and Trimmer 2005).
In the bottom-up approach, one begins with genomic and proteomic data and information to create a model of a system (Noble 2002). A variety of academic institutions and organizations
are working on this approach (Michelson, Sehgal, and Friedrich 2006). These groups are focused on such topics as cellular signaling, cellular systems, and types of cells (e.g., immune cells, tumor cells). Michelson (2003) points out that “One drawback of this approach is that whenever a new component or connection is discovered, the entire model must be reconfigured.”
are working on this approach (Michelson, Sehgal, and Friedrich 2006). These groups are focused on such topics as cellular signaling, cellular systems, and types of cells (e.g., immune cells, tumor cells). Michelson (2003) points out that “One drawback of this approach is that whenever a new component or connection is discovered, the entire model must be reconfigured.”
Stay updated, free articles. Join our Telegram channel
Full access? Get Clinical Tree
