Computational Modeling Core (C)

CORE C – Computational Modeling Core The computational modeling core will provide computational model development and model-based data analysis tools for all the Projects in the Center. A central approach of this Center, and of much work in computational psychiatry, is the use of formal computational models to quantify otherwise abstract functions and connect them to basic neural and cognitive mechanisms. All of the Center's projects concern a class of computational models known as latent cause inference, which describe how humans and animals cluster their experiences so as to identify different underlying contexts in which different rules apply. Such inference has important effects on learning and decision making, and we hypothesize that dysfunction in these processes is implicated in numerous mental illnesses. A key approach for formalizing and testing such hypotheses across all our Projects is estimating the free parameters of the latent cause model: a set of interpretable “knobs” that directly control the behavior of the model. Thus, from an individual's behavior or brain measurements, we can determine the model parameters that best explain the data, and test whether these differ in psychopathology. To support this approach, the Computational Modeling Core has three aims. The first is to develop a single, uniform latent cause model appropriate to the tasks and experimental measurements across all the Projects. The second aim is to build data analysis tools to fit this model to multivariate, multimodal experimental datasets. This involves estimating the model's parameters both for each individual and, using hierarchical modeling techniques, at the group-level and for condition averages. These estimates allow us to conduct statistical tests such as whether the model's parameters are different between healthy and clinical groups or whether the parameters change in a graded fashion with symptoms. Our third aim is to develop tools to estimate graded dimensions of psychopathology (e.g., depression, anxiety) from self-report psychiatric symptom data collected in the Projects. For this, we will use modern hierarchical factor analysis techniques to characterize accurately the separate dimensions of illness, while coping with high levels of comorbidity between them. By concentrating modeling infrastructure in a single Core, all Center Projects will benefit from a unified model that was constrained by data from multiple tasks. The shared model and model-fitting tools will allow easy comparison of results across Projects, and cross-Project insights to emerge. By implementing uniform statistical best practices for data analyses, this Core’s work will also promote the rigor and reproducibility of all the Projects.