Linking transdiagnostic biomarkers to credit assignment impairment in Humans
Abstract
Transdiagnostic traits, such as impulsivity, anxiety, compulsivity, and dependence, are increasingly recognized for their relevance in explaining psychiatric disorders. These traits often cut across traditional diagnostic categories and are thought to be more directly related to underlying cognitive and neural mechanisms. This study investigates how individual differences in reward expectation and belief updating, key components of reinforcement learning (RL), relate to these transdiagnostic traits. Participants performed a probabilistic learning task in which they made predictions and updated their beliefs based on reward feedback. Using general linear models (GLMs), we quantified the contribution of various cognitive processes to reward-based learning, including belief persistence, cue confidence, and prediction error. In addition, factor analyses of psychiatric questionnaires revealed latent traits, with delay discounting emerging as a particularly salient dimension. Machine learning models were applied to assess the predictive utility of learning parameters for classifying high and low scorers on psychiatric traits. Findings demonstrate strong links between delay discounting and impaired learning processes, especially reduced belief updating and increased rigidity in beliefs, providing new insights into decision-making deficits associated with impulsive behaviour. These findings underscore the potential of computational approaches for understanding the cognitive underpinnings of psychiatric vulnerabilities.