Hi Andy, Al and others,
There may be pragmatic reasons for limiting the levels of scale
complexity in a model (ease of communication and code maintenance), but
I have never seen arguments for a scientific basis for this view. Thus,
I see no problem in modeling some processes at the level of gene
expression if that clearly is required. Phenology is one example where I
expect we need something nearer to gene-scale. We wouldn't try to model
the 200+ genes involved in flowering but would want to describe
interactions among the 10 or so genes determining ecotypic differences.
And if you want to simulate wheat production across the US, Europe and
Asia, you should consider that range of genetic diversity. Our latest
effort in this area just came out in Crop
Science:
White, J.W., M. Herndl, L.A. Hunt, T.S. Payne, and G. Hoogenboom.
2008. Simulation-based analysis of effects of Vrn and Ppd loci on
flowering in wheat. Crop Sci 48:678-687.
Rather than establish a priori criteria for levels of scale in a model,
a more rigorous approach would be to create complex sub-models and then
test whether that complexity improves overall simulation. In a sense,
this is what happened with early efforts to model light interception.
Attempts were made to model whole canopies, leaf by leaf, but a
compromise of modeling two classes of leaves, sunlit and shaded, proved
to work quite well.
People who are interested in gene-based modeling and application of
genomics to modeling may want to monitor the iPlant initiative, which
proposes to use information technology to solve major problems in plant
biology. Several people are forwarding the argument that iPlant should
work on the phenotype to genotype problem using simulation models. For
more info see:
http://www.iplantcollaborative.org/about-ipc

Best regards,
Jeff White