Leveraging computationally derived measures of individual differences in learning and decision-making to predict psychiatric diagnosis, symptoms and changes in symptom severity across time - Leveraging computationally derived measures of individual differences in learning and decision-making to predict psychiatric diagnosis, symptoms and changes in symptom severity across time. PI: John P. O’Doherty PROJECT SUMMARY The goal of computational psychiatry is to gain knowledge about underlying neurocomputational processes that underpin psychiatric disorders and to leverage this knowledge for improving diagnosis and treatment. A key step toward achieving this goal is to develop measures of individual differences in computations obtained from a single individual that are reliable, robust and meaningfully relevant to psychiatric dysfunction. In order to attain these objectives, it is essential we substantiate relationships between candidate computational mechanisms and diagnostic categories, symptom dimensions and treatment outcomes. In the present proposal, we utilize a computational assessment task battery (CAB), designed to measure individual differences across a multidimensional array of computational processes. We aim to separate three different variance components contributing to variability in computational parameter estimation: occasion-related variance due to incidental day to day changes in task performance, state-dependent variance that is related to meaningful variation across time in the underlying computations within an individual, and trait-related differences pertaining to stable individual differences in computations across individuals. To accomplish this, we will first implement repeated assessments using this battery across a 1-year interval within an on-line sample, and use hierarchical Bayesian modeling to separate the effect of occasion, state and trait-related variance on these parameter estimates. We will then relate these variance components to diagnostic categories, symptom dimensions and symptom severity measures in a diverse cohort of psychiatric patients (mostly with depression, anxiety and OCD) recruited in Southern California. Finally, we will track the relationship between the computational parameter estimates and changes in symptoms across time in a subset of these patients. We hypothesize that overall diagnosis will be best predicted by trait variance components, while current symptom severity will more closely relate to state-related variance in parameter estimates. Our proposal promises to significantly advance understanding of how to reliably extract diagnostically relevant computationally-derived measures of cognitive phenotypes that could eventually be migrated to the clinic.
