The design of simulation languages for systems with multiple modularities

Biological systems exhibit several characteristics that are not shared by human-engineered systems: there are often no clear module boundaries (or there are several module boundaries, depending on the question being asked); individual parts often serve multiple roles, depending on the behavior being studied; and system characteristics often vary from individual to individual.

Conventional simulation languages, however, do not cope well with this non-modularity. We have developed a theory of the design of simulation languages for such systems, and partially verified it in one case study. We separate the notion of "structure" S of a system from the "behavior" B of its parts. We allow multiple versions of both the structure (S₁, S₂, ...) and the corresponding behavior (B₁,[S_i], B₂ [S_i], ...) of each part S_i. The different structures or behaviors might be alternative theories, or abstractions of each other, for example.

We also have theories of how to interpret the simulations produced by (S_i' B _i [S_i]) pairs. One goal is to extract "design" information, i.e., explain how the system solves problems. Another is to test the effects of alternative models of behavior B₁ [S_i], B₂ [S_i], ... for the same structure, or the effects on behavior due to alterations in structure. A third is to judge the relative degree of consistency between various (S_i , B_i [S_i]) pairs.

We exhibit how these ideas apply to the motor nervous system of the nematodes C. elegans and Ascaris suum. We also provide arguments that this kind of simulation methodology is also applicable to engineered artifacts, such as systems where parts must serve multiple roles.