Data-driven approaches to biological psychiatry: multimodal data and machine learning in the study of psychiatric disorders

Abstract

This thesis adopts a data-driven, multimodal approach to explore the complexity of psychiatric disorders, emphasizing the integration of diverse datasets to achieve a deeper understanding of these conditions. It illustrates this methodology across several chapters, each focusing on different aspects and biomarkers relevant to stress-related psychiatric disorders. Chapter 2 reevaluates the effects of intranasal oxytocin through a multimodal lens. Chapter 3 connects cortisol's in-vivo effects with brain responses and molecular architectures. Chapter 4 introduces a multimodal machine learning model to predict remission in depressive disorder, aiming for both accuracy and explanatory power. Chapter 5 examines the concept of resilience as a discrepancy between expected and observed mental health post-stress, assessing its predictive value for mental health outcomes. Finally, Chapter 6 explores a novel genetic risk score for stress-related disorders, testing its utility and applicability in various contexts. Through a comprehensive and data-centric approach, the thesis aims to uncover new insights into the etiology and treatment of psychiatric disorders, highlighting the importance of considering their multimodal nature.