The overarching goal of computational psychiatry is twofold: to enhance the mechanistic understanding of maladaptive decision-making and to develop precise, dynamic phenotypes that capture individual variability across psychiatric disorders. The curre...
The overarching goal of computational psychiatry is twofold: to enhance the mechanistic understanding of maladaptive decision-making and to develop precise, dynamic phenotypes that capture individual variability across psychiatric disorders. The current thesis aims to present three studies that have aimed to contribute to these goals and bridge computational neuroscience and clinical science. In this thesis, I used model-based decision-making frameworks, mainly including reinforcement-learning (RL) and introduced three empirical studies to address these aims across distinct clinical contexts. Study 1 develops computational markers for predicting treatment outcomes in nicotine dependence by integrating adaptive experimental design optimization (ADO) with ecological momentary assessment (EMA). This approach enables efficient, real-world tracking of delay discounting and ambiguity tolerance, identifying early behavioral predictors of cessation success. Study 2 investigates the computational and neural mechanisms of context-dependent reinforcement learning in addiction, combining model-based fucntional Magnetic Resonance Imaging (fMRI) with an MR-compatible nicotine vaping device. Findings reveal that nicotine reinforcement shifts learning toward frequency-based processing accompanied by dorsal striatal activation in the brain. Study 3 examines how anxiety and depression exhibit distinct temporal integration profiles in probabilistic learning: anxiety shortens the learning window, emphasizing immediate feedback, while depression attenuates this effect, reflecting sluggish adaptation. Together, these studies demonstrate how hybrid computational approaches—linking behavioral modeling, neuroimaging, and ecologically valid data collection—can capture the temporal, contextual, and mechanistic dimensions of psychiatric dysfunction. The thesis advances computational psychiatry toward precision phenotyping, showing that quantifying how individuals learn from feedback and context provides mechanistic insight into clinical conditions such as addiction, anxiety, and depression, and offers pathways for individualized intervention.