Simbios Talk by Ajay Seth, University of Texas at Austin, Jan 10, 2006

Title: A Reverse-Engineering Framework for Modeling Biological Systems.

Abstract:
Biological systems offer unique engineering challenges because they do not behave like ideal mechanisms, circuits, reactions, etc. Moreover, they are often collections of multiple processes operating in different engineering domains as well as on varying spatial and temporal scales. Unlike other engineering disciplines, bioengineers do not have the luxury of specifying their own building blocks from first principles and assembling them into a functional system. In contrast, an organism/organ/tissue/cell represents an integrated system, where the number of components and their functions may be unknown. Biological scientists and engineers, therefore, must, to some degree, “reverse-engineer” the system to understand how it works, and translate that understanding into models in order to predict behavior. However, it is a very challenging task to bridge the gap between formulated dynamics and observed performance especially when there are unknown influences/inputs that dictate how the system will react/respond. Vast amounts of computing resources and intractable amounts of time are often required for search techniques (optimizations) to resolve for unknowns that can vary with time. Since the goal is to have models that will reproduce observations, we asked ourselves if empirical data could be used more effectively to both define the model and determine unknowns (parameters and inputs) that produce more realistic simulations. The central idea is that empirical data does not have to be employed secondarily for comparison purposes a posteriori, but can be used primarily to guide a model and to determine unknowns systematically. In fact, mechanical and industrial engineers have developed and applied elements of control theory for several decades to determine the required actuator inputs and configurations for machines to perform desired tasks.

Similar reverse-engineering techniques can be applied to any system. The basic system requirements are: 1) a hypothesis of the system dynamics expressed as differential equations, 2) a set of unknowns as system inputs (i.e. controls), 3) a set of outputs that correspond to experimental observations and 4) empirical data. Based on control theory, we have developed a reverse-engineering framework for simulating musculoskeletal systems in a way that leverages experimental observations to determine individual muscle forces, which cannot be observed or measured directly. Already, this framework has eliminated much of the need for “brute force” methods, like large-scale optimizations, to solve for unknown muscle forces, and, have reduced computing costs by three orders of magnitude. Such dramatic reductions in computing time and resources allow for many more simulations to be run in a much shorter timeframe. This, in turn, enables models to be developed iteratively– at each step assessing how well the current model reproduces experimental observations and providing the opportunity to adapt the model. This is a straightforward approach to identifying the source of discrepancies between simulated and empirical data and to attribute these differences to specific modeling simplifications and assumptions. These techniques can be applied to the musculoskeletal system as a whole and to its individual components/subsystems so that models can be validated in greater detail and at various levels of complexity. By demonstrating the advantages of this framework in the analysis of a musculoskeletal system we hope to encourage its application to other complex biological systems ranging from protein dynamics to cellular interaction to artificial organs.