Applying Simulation Technology to the Life Sciences
Menlo Park, California
Although simulation is considered a powerful, well accepted approach to understanding the behavior of complex engineered systems, it is not held in the same high esteem in the life sciences. The many differences between the fields of engineering and biology are the results of long-standing different traditions reinforced by undergraduate and graduate degree programs. Engineering embraces the principles of design, systems, and differential equations. By contrast, the life sciences embrace the principles of observation, reductionism, and classical statistics. Clearly, engineers and life scientists have very different ways of looking at the world that present both opportunities and challenges for cross-disciplinary groups engaged in the simulation of biological systems.
SIMULATING ENGINEERED SYSTEMS
For the most part, the components of engineered systems are well understood, their characteristics specified. The behavior of the integrated system, however, is not well understood. Simulation is used to move from component specification to system behavior. Based on iterative testing of changes in component specification, component designs can be refined to generate the desired behavior for the integrated system. This “bottom-up” process is primarily iterative optimization. The use of extensive simulation in Boeing’s design of the 777 aircraft, for example, has been well documented in the press.
In some instances, this process is reversed to move from system behavior to component specification. Commonly referred to as “reverse engineering,” this process has been used to determine the design of an industry or military competitor’s system. Reverse engineering was used, for example, to clarify infrared-
seeker technology for antiaircraft missiles during the Cold War. The reverse engineering “top-down” approach is characterized by uncertainties about component specifications that could produce the observed system behaviors, especially when the system has only been partially observed, either with respect to internal states or temporally. One can think of possible component specifications as hypotheses to clear up these uncertainties. Simulation provides a means of testing these hypotheses.
Both bottom-up and top-down approaches are used in simulating biological systems. Each approach is motivated by different trends and is focused on different applications.
BOTTOM-UP SIMULATIONS OF BIOLOGICAL SYSTEMS
Many groups, particularly in academia, are pursuing the bottom-up approach to simulation. Some of this research is in areas where the “first principles” of the underlying processes have been well characterized, such as the physicochemical processes involved in fluid flow, tissue oxygenation, and basic metabolic pathways. Simulations in other areas, such as attempts to simulate the logic underlying the behaviors of bacteria, cells, and tissues, are based on less of a first-principles foundation. The goal of bottom-up simulations is often to elicit “emergent behaviors” that demonstrate how a relatively simple network of nonlinear components can give rise to very complex behaviors. Simulations of action potentials via fluctuations in ion channel function, which date back to the early 1950s, have yielded insights into cellular electrophysiology.
The advent of molecular biology in the late 1980s provided entirely new technologies for perturbing and observing intracellular processes. Today, academic groups around the globe are developing complex simulations of intracellular processes, but bottom-up simulation is coming up against significant limitations, particularly the problem of observability. Put simply, one can think of two levels of data related to the observability of a biological system: (1) structured-level data, and (2) dynamic-level data. Structural-level data provide information sufficient to affirm the existence of a pathway and describe how it fits into a larger network of pathways. Dynamic-level data provide information to describe quantitatively the nonlinear dynamics of that pathway in the multiplicity of boundary conditions specified by the surrounding network.
Molecular biology has provided a wealth of structural-level data for many systems but dynamic-level data for only a few cellular systems, including cardiac cells. The scarcity of dynamic-level data is likely to continue for some time. The rate at which structural-level data are being generated far surpasses, and is accelerating, relative to the rate at which dynamic-level data are being generated. In the absence of the latter, attempts are being made to use bottom-up simulations to support multiple hypotheses. Managing this gap is a key challenge for bottom-up simulation.
TOP-DOWN SIMULATIONS OF BIOLOGICAL SYSTEMS
The key advantage to top-down simulations of biological systems is that the process begins with the end, the full envelope of functional (phenotypic) behaviors of which an integrated biological system is capable, in mind. Top-down simulations typically begin with simple representations of the dynamics of the system phenotype, because these data are much more prevalent than data on internal states, and therefore, provide a firm foundation for guiding the development of the simulation.
These initial simulations, however, cannot fully reproduce the observed phenotypes. Therefore, more detail is added to more closely reproduce the phenotype envelope. The implementation of these details is constrained by lack of these data, which typically become like more structural-level data as the components of the simulation become more detailed. As the simulation goes deeper, it becomes increasingly necessary to explore multiple alternative hypotheses.
Top-down simulations follow the principle that simulations should be “as simple as possible, but no simpler.” The addition of detail is motivated by testing alternative hypotheses to explain key phenotypic behaviors. By contrast, in bottom-up simulations the detail is motivated by the availability of data.
The several groups in academia that are pursuing the top-down approach at the cellular or tissue level typically have researchers who are both developing simulations and collecting data. This creates a tight iterative loop for the formulation and testing of hypotheses in silico and the confirmation of hypotheses in vitro. These top-down simulations help researchers design efficient, incisive experiments and then provide a framework for interpreting experimental results in the larger context of the integrated system. Entelos, Inc., uses this same top-down approach at the clinical (human) level. Collaborative relationships with our pharmaceutical partners provide the empirical capabilities that complement our simulation technology and the expertise in in silico research.
THE CENTRAL ROLE OF SIMULATION IN BIOLOGY
The hypothesis-data cycle just described is central to the advancement of science. The physical sciences have benefited from both theoretical and empirical research made possible by the common language of mathematics since their inception. By contrast, the classical statistics used in the life sciences help researchers describe that certain phenomena occur but not how they occur. Prior to the advent of molecular biology in the 1980s, research in quantitative physiology had made significant progress using engineering-like approaches to understanding biology. Since then, however, molecular biology, with its high-throughput data collection (rather than formal hypotheses of integrated biological systems), has taken center stage.
Until recently, empirical technologies have completely outpaced technologies
for managing hypotheses because there is simply too much data for researchers to keep in their heads. This is why simulation, particularly top-down simulation, can play a central role in biology. With this kind of simulation, hypotheses of biological function (or dysfunction) can be made explicit and open to critique and review. Top-down simulations can also address fundamental issues of complex biological systems that cannot be managed in mental models, including redundancy, feedback, and multiple timescales.
Simulation also has risks, of course. Black-box simulations cannot serve as vehicles for dialogue and inquiry; for that they must be made into “clear cube” simulations. Simulations must be considered as a key technology in the discipline of in silico research and not accepted on blind trust. Simulations also have organizational risks. Top-down simulations have the potential to show that some research projects are likely to be ineffective in the clinic. Although this would be tremendously valuable to the organization, individual researchers will feel challenged.
In the past 10 years, Entelos has addressed these issues through integrated development of methodologies for in silico research and the development of proprietary software technology. All of our researchers, engineers and life scientists, scientific advisors, and software engineers have contributed to the advancement of our top-down simulation.
THE FUTURE OF SIMULATION TECHNOLOGY IN THE LIFE SCIENCES
Many projections have been made concerning the impact of research and health care in the postgenomic era, including predictions of innovative new therapies and personalized treatment regimens based on individual genetic factors. All of these projections depend on an understanding of the role of specific genes and proteins in integrated human physiology. For the reasons described above, top-down simulation will be pivotal to making these possibilities a reality. Research in the life sciences will change accordingly, with dual, yet integrated, communities for theoretical and empirical research much as we find in the physical sciences. With an impressive array of high-throughput empirical technologies, the paradigm will shift from a shotgun, “data for data’s sake” paradigm to fill databases for later investigation to a focused, iterative, “data to confirm detailed top-down hypothesis” paradigm. All of these changes will be led by industry rather than academia because the pharmaceutical and health care industries are facing acute challenges that are motivating them to advance in silico technologies, methodologies, and applications.