Reverse Engineering Depression: Data-Driven Discovery of Causal Dynamical Models

Resumen de los autores

Recent research focusing on formal models of psychiatric disorders has provided a basis for understanding mechanistic relationships between symptoms. Using a complex systems approach, dynamical systems and statistical methods can be used to analyze the interacting parts of the whole psychological framework. Nonlinear differential equations have been of particular interest as they lend themselves well to exploring how symptoms change over time, discovering tipping points, or characterizing attractor states. Thus far, research of this kind has focused primarily on a priori definitions of these disorders using simulated data. While these simulated models have been extremely insightful for discerning the fundamentals of psychological phenomena, much less research has been devoted to aligning these systems with real-world data. We apply a data-driven approach, which allows researchers to deduce systems of differential equations from psychological datasets. We exemplify this method using a theoretically defined model for depression and a dataset measuring passive sensor information, e-communication data, and psychological symptom assessments. In this article, we demonstrate how these variables can be integrated into the differential equation discovery framework and methods for parameterizing nomothetic or idiographic models for the disorder. Overall, we found that empirically determined equations aligned well with our predefined theory of depression, suggesting that a combined empirical and theoretical approach may complement each other in further model development. We also found that personalization offered incremental benefit over global models, suggesting that an idiographic or combined idio-nomo approach may improve further formal model advancement by more explicitly considering individual differences in model parameterization.

Áreas y palabras clave

Cómo citar (APA 7)

Engels, S., & Curtiss, J. (2026). Reverse Engineering Depression: Data-Driven Discovery of Causal Dynamical Models. https://doi.org/10.31234/osf.io/xqpgc_v2