Neural computations of learning, decision-making and memory

Most of our behavior consists of learning about the environment, choosing between options that we learned about, and storing the learning in memory for future use. These processes rely on a set of basic computations, including estimations of value and uncertainty and calculations of prediction errors. Prior research has

identified neural correlates of these computations, and linked aberrations in these computations to maladaptive behavior and psychopathology. This research, however, typically focused on one process (e.g. learning, decision-making or memory), within one domain (e.g. rewards or punishments). Our goal in this application is to

characterize core computations of value, uncertainty and prediction error, across processes and domains, within the individual, and to identify associations between these computations and symptoms of anxiety and stress- related disorders. We plan to do this in a behavioral study with 1000 participants (Aim 1), and a functional MRI

study of 100 participants (Aim 2), including both men and women from the general population. We will employ a well-validated decision-making paradigm, together with a novel naturalistic game-like experimental paradigm, which combines a passive-learning stage, a decision-making stage, and a memory stage administered on a

second day. Our design will include explicit measures, and latent variables which will be derived from computational modeling. In Aim 1, we will characterize the computations behaviorally, by constructing a distribution for each measure, based on the whole sample, and characterizing each subject based on their

location in this distribution. This will allow us to estimate similarities and differences between computations of the same measure (e.g. value) across processes (learning, decision-making and memory) and domains (rewards and punishments). In Aim 2 we will characterize the computations neurally, by examining activation and

connectivity patterns. We will estimate to what extent these patterns for each measure (e.g. value) in one process (e.g. learning) predict the patterns for the same measure in other processes (e.g. decision-making and memory) and to what extent representations in the reward domain predict those in the punishment domain. Aim 3 will

classify participants based on the behavioral and neural characteristics, and identify associations between these classifications and clinical symptoms of anxiety and stress-related disorders. We hypothesize that different psychopathology symptoms are associated with the type and magnitude of alterations in basic computations.

Together, the proposed studies will inform our basic understanding of the computations of value, uncertainty and prediction error, and will unravel links between these computations and clinical symptoms, in a dimensional manner, consistent with the RDoC approach. In the long term, we expect the methodology developed as part of

the proposed studies to be useful in characterizing neurobehavioral differences in a variety of conditions beyond anxiety and stress.