Multimodal Representation Learning for Parsing Biological Heterogeneity in Psychiatric Neuroimaging.
Review
Overview
abstract
For decades, psychiatric neuroimaging has searched for biomarkers of depression and other disorders, but they remain elusive in clinical practice. While the last five years have seen rapid progress, other large-scale correlative studies have found only small, unreliable links between brain measures and clinical symptoms. Growing evidence suggests that such limitations are not just about sample size but depend critically on how models represent data. This review traces a recent shift away from univariate methods to multivariate/multiview approaches that learn more effective representations of biological and symptom measures by flexibly learning multimodal latent representations. We first review how linear multiview embedding methods have revealed reproducible biological depression subtypes, but do not perform well in small samples or samples enriched for mild symptoms. We then consider newer work exploring more sophisticated representations for neuroimaging data, including deep-learning and graph-based representations, and multimodal extensions that uncover complex latent patterns that single-modality studies miss. We then review recent developments in "foundation" models which, once trained on large corpora, can "transfer learn" readily to small clinical cohorts, potentially bringing the advantages of large-scale learning to small, privacy-limited data. Finally, we highlight emerging representation tools that treat the brain as a dynamic, stateful multivariate process. Taken together, these advances point to a future in which the value of neuroimaging will be determined not just by ever-larger sample sizes but also by data quality and by how well our algorithms capture the distributed, multimodal, and evolving nature of psychiatric disorders. Linking psychiatric symptoms and behaviors to their underlying neurobiological mechanisms, and ultimately discovering biomarkers for diagnosis and treatment, is a primary goal of psychiatric neuroscience. Despite rapid progress in the past decade, clinically useful biomarkers remain elusive for two related reasons: psychiatric diagnoses are heterogeneous, complicating efforts to model them as unitary conditions; and univariate effect sizes in psychiatric neuroimaging are typically small and difficult to replicate. These challenges are especially apparent in depression, a primary focus of this review. Major depressive disorder (MDD) is not a unitary disease but a clinical label for a constellation of symptoms that can combine in >200 distinct ways under current diagnostic criteria. Distinct mechanisms underlie divergent presentations, and conversely, different mechanistic pathways can yield superficially similar symptom profiles. The same pathophysiological mechanism may also manifest differently across individuals, shaped by environmental factors, neurodevelopment, and other moderators. We view this heterogeneity not only as a challenge but also as a scientific opportunity. Parsing individual differences in symptoms, neurobiology, and treatment response has the potential to sharpen effect sizes and reveal mechanistically distinct subtypes. Rather than seeking one-size-fits-all models, this approach aims to define biologically coherent subgroups and brain-behavior dimensions that explain individual differences, and ultimately support mechanistic, personalized treatments and improved outcomes.
