Viewpoint Aggregation
While the biological systems themselves are the ultimate “reference point” for any examination of their processes, it is our understanding of those systems that is the true subject of bioscientific inquiry. Consequently, our approach is to quantitatively compare and contrast multiple points of view as each provides its own unique insight into the system. While the viewpoints for a given system are unique to that system, they typically can be categorized as “top-down,” “bottom-up” or “middle-out.”
Top-down viewpoints tend to be teleological in nature as they hypothesize why a system does what it does. Thus, a top-down model of a system might be comprised of statements positing the purpose of various structures or processes within the model. “The purpose of the heart is to pump blood” would be an example of a top-down hypothesis/model.
Bottom-up viewpoints tend to be mechanistic/first principles-based descriptions from which higher-level behavior is expected to emerge. Thus, bottom-up models try to explain how a system does what it does. “The heart is a collection of contractile elements organized into chambers with one-way flow ports” would be an example of a bottom-up model.
Middle-out viewpoints represent experimental/clinical data. Thus, middle-out viewpoints are ideally devoid of assumptions, hypotheses and conclusions and instead encompass what the system does. “The heart receives neural signals, it contracts periodically and blood flows through it.”
Importantly, characterization of these viewpoints is context-dependent. That is, one researcher’s top-down conceptualizations are another researcher’s, bottom-up mechanisms. For example, the “contractile elements” mechanisms of the heart function researcher would be top-down conceptualizations of the cardiac cellular electrophysiologist.
Viewpoint aggregation is our method of quantitatively comparing these differing perspectives with the ultimate goal of reconciling them to one another. That is, the why, how and what should all represent the system equally well. Our primary tool for reconciling these disparate viewpoints is “relational analysis.” Relational analysis seeks to look past the model/experimental values and instead capture the relationships between those values. This approach is based on our theory on the fundamental nature of biological systems called Heuristic Emergence via Dimensional Restriction and has been shown to be quite effective in multiple systems.
Relational Analysis
A central feature of experiment-based biological research is the examination of the relationships exhibited by the system. Thus, statistical t-tests and linear regression are the hallmarks of experimental literature. Ironically, most model-based biological research focuses on values rather than relationships. A model is considered “validated” if it produces the desired output values. However, in systems with changing dimensionality, there is a very real potential for such a model to be non-unique. That is, the model may merely be one of many possible descriptions of the system. The reintroduction of relationships as part of the validation can ameliorate this problem to some degree in that a vastly richer set of validation criterion can be examined (N measures of output become potential N2 outputs and relationships).
We have developed a set of practical methods based on examining the presumed “intrinsic” relationships exhibited in a dataset (experiment or model derived) that we collectively call relational analysis. When used in conjunction with our viewpoint aggregation methodology, relational analysis is a powerful modeling tool that can dramatically speed model development, and generate predictive assessments of posed hypotheses as the “landscape” produced by relational analysis provides the means for the apples-to-apples comparisons between the various perspectives.
Selected References
- Mitchell CS, and Lee RH. Output-based comparison of alternative kinetic schemes for the NMDA receptor within a glutamate spillover model. J Neural Eng 4: 380-389, 2007.
Mechanistic and Conceptual Modeling
While high-minded discussions of complexity theory, analytical techniques, and biological principles are quite interesting, as “practical neurophysiologists” it is important to remember that the details are central in biological systems. This translates into a need for deep understanding of our chosen systems and detailed mechanistic and conceptual modeling. This in turn, translates into immersion into the literature and community of that system. Thus, our modeling techniques augment rather than replace traditional modeling approaches to speed model development and strengthen analysis and prediction.
We have developed cutting-edge models (or more accurately sets of models) of each of our chosen systems each based on both first-hand and published experimental data as well as proposed theories of system function. In accordance with our multiple viewpoint approach, we highly emphasize management and categorization of the embedded principles, assumptions, hypotheses and experimental data that form the bases of these models. Models are implemented in a variety of computer platforms based on the model needs as well as the need for customization and simulation automation.
Selected References
- Mitchell CS, Feng SS, and Lee RH. An analysis of glutamate spillover on the N-methyl-D-aspartate receptors at the cerebellar glomerulus. J Neural Eng 4: 276-282, 2007.
- Shapiro NP, and Lee RH. Synaptic amplification versus bistability in motoneuron dendritic processing: a top-down modeling approach. J Neurophysiol 97: 3948-3960, 2007.
High-Performance Computing
In any modeling endeavor, there is the inevitable point of needing more computational power. To this end, we explore alternative computing platforms. For the past several years, the best of these alternatives has been reconfigurable computing devices called FPGA’s (field-programmable gate arrays). These devices are a mainstay of the high-end electronics industry and are ideally suited to simulating the systems of differential equations that underlie most biological models. We have repeatedly demonstrated performance gains of 20 or higher compared to general-purpose CPU-based simulation at a comparable cost. However, they are not easy to program. Consequently, our efforts have focused on facilitating FPGA-based simulation. Our latest endeavor is a “simulation compiler” that automatically generates the FPGA-compatible code directly from the differential equations (patent pending).
Selected References
- Weinstein RK, Reid MS, and Lee RH. Methodology and design flow for assisted neural-model implementations in FPGAs. IEEE Trans Neural Syst Rehabil Eng 15: 83-93, 2007.
- Weinstein RK, and Lee RH. Architectures for high-performance FPGA implementations of neural models. J Neural Eng 3: 21-34, 2006.
- Graas EL, Brown EA, and Lee RH. An FPGA-based approach to high-speed simulation of conductance-based neuron models. Neuroinformatics 2: 417-436, 2004.