Fig. 1. Illustration of the validity appraisal guide for computational models of psychiatric disorders. Abbreviations and symbols are shown in the legend panel.

Predictive validity (Panels A₁ and A₂): The presence of predictive validity requires identifying distinct features (here F₁ and F₂), each of which is specifically explained by distinct models M₁ and M₂. One must show that there exists a real-world transition (such as medication use) that results in the transition from feature F₁ to F₂, and that this can be adequately modeled by a transformation of model M₁ into M₂. Face validity (Panels B₁, B₂, C₁, C₂): To establish face validity, one must identify features that characterize a target condition (such as bipolar disorder; here Condition A), denoted F₁^A, F₂^A, … , F_N^A. Ideally, features that characterize a relevant comparator, Condition B, should also be identified (F₁^B, F₂^B, … , F_K^B). If model M has face validity, then it should be able to explain as many features of condition A as possible, while not explaining features of condition B. Finally, if model M explains some feature F₁, then it should not explain mutually exclusive features F₂. Construct validity (Panels D₁, D₂): To establish construct validity, one must identify the components of a natural system, such as a biochemical pathway or neural circuit, and establish that the functioning of that system explains some feature(s) F. A model system has construct validity if it is specified at a level of abstraction such that individual components and interactions are homologous to those present in the natural system.