Agent Based Modeling (ABM) is a synthetic or constructive modeling style where the primary objective is to construct collections of composite computational actors whose collective behavior generates phenomena similar to the model's referent (the system being studied).
|
|
2009-11-05
| New paper: At the Biological Modeling and Simulation Frontier has been published in Pharmaceutical Research, Volume 26, Number 11 / November, 2009. This paper delves deeply into modeling and simulation methodology. But it's not one of those papers filled with box-and-arrow figures that yaps on and on about useless abstractions! No. It starts from the practical task of simulating 5 various fine-grained biological systems and builds out from there, through our oft-cited co-simulation method, and out to a clear and, I think practical, explanation of the roles for deduction, induction, and abduction. The fundamental threads of symbolic grounding and model robustness in this paper grew out of the previous paper in Complexity.
|
2009-08-08
| More Good news! Our paper: Evaluating an Hepatic Enzyme Induction Mechanism Through Coarse- and Fine-grained Measurements of an In Silico Liver has been published in Complexity Vol. 14/No. 6. We even landed the cover, thanks to Tony's fantastic artwork! This paper has two main prongs: 1) it demonstrates a concrete example of Petroski's principle and the scientific method, in general, and 2) measuring a system, in this case the simulation and its referent, from multiple aspects, one coarse and one fine. The trick is that the referents (isolated perfused rat livers) are only measured at the coarse grain. Although we couldn't add enough detail (due to space restrictions) to explicitly lay out how indirect validation can be done at the fine grain, the article discusses the methodology for such. In addition, we describe a nice counter intuitive result typical of ABMs of complex systems where decreasing the likelihood of a metabolic event lead to increased extraction because the liver compensated by creating more enzymes.
|
2008-10-17
| New paper on the In Silico Liver published in Complexity. In this paper, we demonstrate how synthetic modeling (including agent-based models) should
be used for scientific research. Often, ABMs are used in the same way
traditional (operations research or control systems style) simulations
are used. Traditional simulation relies fundamentally on the analytic
expressibility of the model. Granted, often times, the mathematics of
these traditional models have to be solved numerically; but the
characterization of the system is still done with continuum
mathematics. And (bear with me, here) that means that the models
express relations between variables. And where do the
variables come from? From either data taken off the referent system or
from a hypothetical assertion: "if we had data, it would follow a
continuous curve (or manifold) like so". Built into such methods is
the requirement to take data off the system. You cannot falsify such a
model without actually performing the experiments on the referent.
With synthetic models, on the other hand, you can, because synthetic
models allow you to explore hypotheses without specifying the variables
in advance. Instead, you specify the mechanisms and can (to some
extent) leave the phenomena you measure (the data, the measure-ments)
open. This allows you to explore intra-model consistency, instead of
assuming such consistency as we do with traditional variable-based
(inductive) models.
In our paper, Evaluating an hepatic enzyme induction mechanism through coarse- and fine-grained measurements of an in silico liver,
we document a set of experiments that are currently infeasible on real
livers. That means there is no data against which to falsify or
validate the model. Yet, we still manage to falsify the hypothesis
reified in our model, thereby eliminating some wet-lab experiments we
might otherwise have performed in order to falsify the model.
The
Holy Grail of modeling and simulation is exactly this: To perform
experiments in silico in order to obviate or intelligently guide
expensive experiments on the referent system. It often surprises me
how often simulation is used in industry and academia with zero
consideration or achievement of this primary objective. So, we're very
happy Complexity accepted this article for publication. It's amazing
how few scientific journals will not publish reports of falsification,
despite falsification being the primary element in the scientific
method.
|
|
|
