What is Computational Psychiatry? An introductory narrative review

The aim of this review is to provide an introductory overview of Computational Psychiatry (CP). CP is a broad, interdisciplinary field that integrates concepts and methods from mathematics, informatics, neuroscience and related disciplines to improve the understanding, prognosis, and treatment of psychiatric disorders. CP seeks to leverage the advances of computational tools to address the intrinsic complexities of psychiatry phenomena and to overcome long-standing limitations of psychiatric research, including the lack of formal theories, the reliance on descriptive and agnostic diagnostic systems, the scarcity of valid biomarkers, the predominance of simplistic outcome measures, limited treatment specificity, and the low reproducibility of many findings.

CP encompasses two complementary strategies. Theory-driven approaches focus on the development of generative models often informed by cognitive neuroscience, that provide formal (mathematical) representation of latent mental processes, within frameworks such as belief updating, decision-making, or reinforcement learning. These models aim to explain how psychiatric phenomena emerge from altered neural computations, thereby offering mechanistic links between neurobiology, behaviour and subjective experience. Data-driven approaches, instead, leverage Machine Learning and other inferential techniques to analyse large, multimodal datasets. This enables the identification of latent structure in complex data supporting tasks such as disease subtyping, risk stratification, prediction of clinical trajectories and estimation of differential treatment response.

Together, theory-driven and data-driven approaches provide complementary tools to generate and test specific hypotheses about psychiatric conditions, while advancing precision psychiatry through the tailoring of interventions to individual etiological, clinical, and pathophysiology profiles. This review illustrates key applications of CP through selected examples, including Reinforcement Learning models of anhedonia, Drift Diffusion Models of evidence accumulation, and contemporary AI approaches such as deep learning and explainable AI. These examples highlight how CP can move psychiatry beyond descriptive symptom frameworks, toward explanatory, predictive, and personalised accounts of mental disorders. While CP holds promise for the development of more precise diagnostic and therapeutic tools, important challenges remain, including issues of generalisability, interpretability, ethical oversight, and integration into routine clinical practice.