I spent the first few days of this Michigan spring in the front lawn reading a book which has sat on my shelf for the better part of two years. While I am only a few chapters in, the book has already presented several interesting points within the context of modeling and data analysis. In “The Model Thinker”, Dr. Scott E Page evokes a plea to reader: employ a multi-model approach to solve problems. Essentially, this boils down to the idea that developing multiple computational or mathematical models to solve a single problem (or many problems) is preferable to creating one model to solve a single problem.
While this might appear initially appear obvious, this approach presents an antithesis to modeling as it is introduced within any single paper in the literature. Most systems biology papers, for example, tend to follow the general format of:
1) Introduce the problem I am trying to solve and why it is important
2) Describe the single model I developed in order to solve this problem
3) The model predicts these solutions to the problem
4) Some of the assumptions we made could be incorrect, but future work could identify that.
Importantly, the design of this representative paper’s presentation is one-to-one, wherein one model is created to solve one problem. Of course, there are exceptions in the field (I am reminded now of the Mark Twain quote, “All generalizations are false, including this one”) and some fantastic papers have been crafted that use multiple mathematical models to compare competing theories. this paper from Melissa Lever in Dr. Omer Dushek’s group on T cell receptor signaling is a favorite example of mine at the moment. However, to my knowledge, these papers represent outliers within the context of modeling and systems biology.
So where do we go from here? The general format of a research paper does not lend itself to exploring the “failed models”. This extends to the experimental, or ‘wet-lab’ world as well. When a scientist writes about a new mouse strain that perfectly recapitulates human obesity, they do not spend time describing all the failed strains. Now, review papers, by definition, collect multiple findings from different models (mice, mathematical, monkey, human, computational), and represent the very few papers which present multiple models in an effort to solve the problem(s) at hand. Therefore, I guess, review papers present the collective intelligence of the field – which is a direct and successful product of being the multi-model thinker that Dr. Page advocates for.
Perhaps, as others have suggested, the scientific paper is an antiquated form of communicating the scientific process and needs updates so that it shows the “exploration” of multiple models prior to final presentation via a paper. Undoubtedly, the strain of mouse that acts as a perfect model for obesity was not the first such model (nor the best under every circumstance) and should probably be presented as such. Likewise, in my own work, we have explored several models before settling on the one that best answered the problem at hand. These other models never make it into the condensed, clean and final version of a scientific paper. Perhaps they should.