Making Sense of Computational Psychiatry
In psychiatry we often speak of constructing “models.” Here we try to make sense of what such a claim might mean, starting with the most fundamental question: “What is (and isn’t) a model?” We then discuss, in a concrete measurable sense, what it means for a model to be useful. In so doing, we first identify the added value that a computational model can provide in the context of accuracy and power. We then present limitations of standard statistical methods and provide suggestions for how we can expand the explanatory power of our analyses by reconceptualizing statistical models as dynamical systems. Finally, we address the problem of model building—suggesting ways in which computational psychiatry can escape the potential for cognitive biases imposed by classical hypothesis-driven research, exploiting deep systems-level information contained within neuroimaging data to advance our understanding of psychiatric neuroscience.
Figure 1. Three “models”: Freudian iceberg model of the human mind, modern neuropsychiatric control circuit model, and dynamical systems model of a control circuit. (A) An example of the typical conceptual models historically grounding psychiatry, Freud’s theories assume that behavior reflects unconscious influences provided by components of the mind, including drives (“id”). (B) Schematic of the cortico-basal ganglia-thalamo-cortical (CBGTC) loop, integrating data from multiple neuroimaging (MR-PET) modalities. Crucially, the CBGTC loop can be treated either as a heuristic (as per the Freudian iceberg mind) in which it functions more as a map than a true model—meaning that it cannot be operationalized to make predictions over time—with new inputs, or it can be treated as a graphical representation of a dynamical system of differential equations as per (Frank, 2005) and the PID control circuit shown in C. (C) The dynamical system model is distinguished by the fact that, when presented with time-series inputs, it produces simulation outputs. These output trajectories can be compared with new data to rigorously assess the model’s validity. Images adapted from (Maia and Frank, 2011) and Arturo Urquizo (CC-BY 3.0).
Figure 2. Circuit discovery. To illustrate how transfer function structure changes with different circuit topologies, we show 3 transfer functions, each of which corresponds to a different kind of “motif,” with series (A), parallel (B), and feedback (C) connections. By using pairs of inputs (u) and outputs (y) to obtain their transfer function (system identification), we can systematically infer circuit topology. A hybrid version uses machine learning to pattern-match data dynamics to the canonical dynamics produced by each motif. This type of data-driven reverse engineering has been used successfully across science and engineering domains, including systems biology (Bongard and Lipson, 2007; Luo et al., 2010; Brunton et al., 2016).
For more information, see:
Making Sense of Computational Psychiatry
Lilianne R. Mujica-Parodi and Helmut H. Strey. International Journal of Neuropsychopharmacology, Volume 23, Issue 5, May 2020, Pages 339–347.
