Our modeling community has built a jumble of models for neural processes at different scales using different abstractions, amenable to different scientific questions and constraints by different kinds of data. In the bigger picture of understanding a multi-scale disease such as epilepsy, determining the compatibility and consistency of those models and the available data poses a huge meta-scientific challenge. Our current approaches to this challenge rely on simulation, toy mathematical models, superficial metadata, and a limited conception of "parameter fitting". I suggest that we will not be able to build useful, large data-driven models of disease that we adequately understand using these approaches. In particular, I will argue that we will struggle to make robust predictions about affecting macroscopic outcomes due to microscopic changes.

I will discuss emerging strategies from various sources across the computational sciences that could change this picture over the next decade, and provide some early prototypical examples of how we could model multi-scale disease mechanisms differently. Using one modest example, I will illustrate a strategy that is leading to a detailed mathematical explanation of the familiar Phase Response Curve (PRC) for a single neuron in terms of underlying ionic mechanisms. The PRC is used in many modeling studies associated with network synchronization despite its poorly understood causal origins. A clearer understanding of this issue will better connect microscopic and macroscopic processes relevant to dynamic, multi-scale diseases.