Click here for our publications.
Click here for a condensed view of this page (research project titles only).
The Center’s research seeks to bridge the conceptual gap between detailed knowledge of the "molecular parts catalog of life" and higher-level “emergent” or "macroscopic" descriptions of cell dynamics and behavior using computer imaging and the most realistic feasible simulations. Research project summaries:
|
|
|
|
|
|
|
Theory finds a role in biology whenever laboratory practitioners find themselves confronted with the task of synthesizing information, acquired in reductionist studies, about the behavior of parts (whether the parts are molecules, genes, cells, or organisms) into descriptions of the function of the whole (where the whole could mean anything from individual macromolecules consisting of monomers, to entire tissues, or even natural populations). This, in essence, is the very nature of the modern biological quest: to comprehend how the self-organizing complexity of life emerges from swarms of material components that obey the laws of physics and chemistry.
Genome projects, DNA microarrays, and other high-throughput techniques are providing an embarrassment of molecular-level riches to cell and developmental biology. As identifying components becomes quicker and easier, it becomes feasible in principle to synthesize these components into whole mechanisms, with constraints and dynamics that emerge from the sum of the parts, but which cannot be predicted from a mere list of those parts. Because even the simplest interesting cases involve networks consisting of dozens of molecular participants, interacting according to non-linear rules and involved in diverse feedback phenomena, it is often beyond human intuition to predict what behaviors, tendencies and constraints might emerge from what recipe of participants.
Using computer simulations, biologists can ask what behaviors to expect from some network of molecules, and test how the predicted behavior depends on the assumptions used to concoct the model. Models, when they work, demonstrate that the molecular facts you know (or assume) can actually generate the higher-level emergent behavior you seek to explain. Especially when they fail, models can provide feedback to experiments by highlighting missing information. Models can suggest what kinds of experiments might reveal details about the mechanism in question, and can predict unanticipated behaviors in response to perturbation.
We prefer agent-based models in which each of many thousands of individual parts is governed by a small system of ordinary differential equations characterizing how it interacts with any/all the other parts. In a cytoskeletal dynamics model, a typical individual agent could be a centrosome or a single kinesin motor molecule, or a segment of a microtubule or actin filament (with the entire microtubule or actin filament comprising tens or hundreds of segments). Since any agent can, potentially, go anywhere and collide with, then interact with, any other agent, a large part of the computational task is to detect collisions. As opposed to the classic continuum model approach in which analytic solutions, or approximations thereto, are possible, these agent-based models require numerical computer solution of their systems hundreds of thousands of coupled differential equations. At first glance this seems a step backward, away from the elegant, conceptually simple approach in which just the process of formulating a classic-form continuum mathematical model distills out a conceptual understanding of what causes the phenomenon under study. It seems to be a step toward a needlessly complicated and computationally expensive kind of modeling involving more arithmetic than thought. But this disadvantage is more than compensated by the principal virtue of agent-based models: the input to such models consists only of low-level facts/assumptions, usually very simple, about how small parts interact. The entire connection between the interactions among the swarm of parts and the systems-level behavior that emerges is the computer solution of the coupled system of ordinary differential equations. The virtue of agent-based modeling is that it bypasses the use of human intuition to distill into an abstract constitutive equation the high-level consequence of many agents interacting. Since human intuition can so easily go wrong, it is a virtue to replace (error-prone) thought by (a great deal of) simple arithmetic. Agent-based modeling is a more direct demonstration that the simple interaction rules by which the myriad parts interact actually can (or cannot) account for what complex phenomena emerge from them.
