Unraveling brain heterogeneity across scales, systems, and diseases

Mapping the progression and heterogeneity of a disease can provide insights into the disease biology and a mechanism for stratifying individuals. Disease progression models can identify the stages of a disease from a dataset of individuals on a common disease trajectory. Clustering tools can identify subtype heterogeneity from a dataset of individuals at a common disease stage. However, commonly within a dataset, there is variability in both disease subtype and stage. This variability confounds the use of disease progression models to identify disease stages and the use of clustering tools to identify disease subtypes. The Subtype and Stage Inference (SuStaIn) algorithm combines ideas from disease progression modeling and clustering to identify subgroups of individuals with distinct progression patterns, simultaneously resolving both progression stages and subtype heterogeneity. In this talk, Dr Young will present the SuStaIn algorithm (PMID: 30323170) and applications to imaging, biomarker, and pathology data, primarily in neurodegenerative diseases but also more broadly in other chronic conditions such as psychiatric disorders and lung diseases. In particular, Dr. Young will focus on recent developments and applications of SuStaIn that use longitudinal data to infer the timescale of disease progression patterns and characterize mixed pathology. 

Disease heterogeneity remains a major obstacle to achieving precision diagnostics. Artificial intelligence (AI) advances have introduced promising approaches to tackle this complexity, particularly by identifying imaging-derived biomarkers capable of predicting disease progression and mortality. Moving beyond traditional unsupervised clustering methods like K-means, Dr. Davatzikos and his team have developed a novel weakly supervised learning framework. This approach leverages generative adversarial networks (GANs) and non-negative matrix factorization methods, in addition to adversarial autoencoders, to model disease trajectories from the healthy control domain to the patient domain, capturing biologically meaningful variance while reducing confounding influences such as demographic factors. In this talk, Dr. Davatzikos will introduce several weakly-supervised AI methods to unravel the neuroanatomical heterogeneity of aging and brain diseases. Among these are Smile-GAN (PMID: 34862382), a model that identifies imaging-derived disease subtypes, and Surreal-GAN (PMID: 39147830), which provides a continuous representation of disease heterogeneity through representation learning. He will also discuss Gene-SGAN (PMID: 38191573), a method that builds upon Smile-GAN by integrating imaging and genetic data to enhance understanding of disease heterogeneity. Beyond methodological innovation, Dr. Davatzikos will highlight the application of these models in diverse contexts, including normal aging and disorders such as Alzheimer’s disease and schizophrenia. This talk will exemplify how advanced AI and imaging techniques can model disease heterogeneity to advance precision medicine. 

Biological heterogeneity presents a formidable challenge in understanding and improving the clinical management of psychiatric disorders. However, the extent of this heterogeneity remains to be investigated at scale or in a pan-disease setting. I will present new data characterizing biochemical and brain heterogeneity in a cohort of 445,374 individuals, considering 9 psychiatric disorders, 30 non-psychiatric chronic medical conditions, and a healthy control group. We find that heterogeneity in psychiatric illness across biochemical and brain phenotypes is 1) comparable to non-psychiatric illnesses, 2) age-dependent, and 3) misaligned with current hierarchical taxonomies of psychopathology. In schizophrenia and bipolar disorder, developmental brain markers (e.g., surface area), as well as glycemic and kidney markers (e.g., cystatin C and creatinine), were most heterogeneous, whereas lipid and hormone traits (e.g., cholesterol and sex hormone-binding globulin levels) were most variable in anxiety and depression. This work provides the first hierarchical classification of human diseases based on biological heterogeneity. Our novel characterizations of biological heterogeneity across a range of psychiatric and medical illnesses lay the groundwork for biological focal points to aid precision medicine in psychiatry.  

Depression often presents with co-occurring physical health conditions, including heart disease, diabetes, and obesity. While dysregulation of the immunometabolic system is posited to underpin several of these physical comorbidities, the prevalence and course of immunometabolic dysregulation in depression is poorly understood and its impact on structural brain changes linked to the disorder is unknown. Using brain imaging and high throughput metabolomics data from the UK Biobank, we comprehensively evaluate cross-sectional and longitudinal immunometabolic profiles in depression, including systemic inflammatory markers and metabolites related to lipid, glucose and amino acid metabolism. Crucially, we find that immunometabolic dysfunction predates illness onset (7 years on average), manifesting a relatively persistent pattern over time of elevated inflammation, upregulated very-low-density lipoprotein and lipids, and downregulated high-density lipoprotein. We also map network-level systemic changes in metabolites in depression, implicating the core role of glycolysis. We show that peripheral immunometabolic dysfunction, particularly elevated inflammation, is associated with brain gray matter atrophy. We conclude that altered lipids and inflammatory markers predate the onset of depression, remain altered throughout the illness course, and explain the severity of brain atrophy. By comprehensively profiling immunometabolic dysfunction in depression and related brain changes, our work highlights the importance of treating chronic low-grade inflammation and altered lipid and glucose metabolism in the disorder.  

The advent of artificial intelligence (AI) has ushered in a transformative era for human brain mapping. Integrating imaging, genetics, and proteomics with advanced computational techniques has opened unprecedented avenues for research, offering a more refined and multi-omics perspective on the brain and related diseases. Building on prior discussions on biological age and imaging-derived disease subtypes, this presentation by Dr. Wen explores the next step: combining AI-driven imaging endophenotypes with additional omics layers, including genetics and proteomics (PMID: 38718880). Dr. Wen will highlight approaches to integrating multi-organ imaging, genetic, and proteomic data, unraveling the genetic and molecular mechanisms of human aging and brain diseases. Two main topics will be discussed, including the multi-organ and multi-omics biological age derived by AI (PMID: 38942983; 38521789) and disease subtypes of four brain disorders (i.e., Alzheimer’s disease, autism, schizophrenia, and late-life depression; PMID: 37662256). The talk aims to shed light on the interplay between AI, multi-omics, and the complex processes driving aging and disease by bridging these data sources. Ultimately, this presentation seeks to demonstrate the potential of AI-powered, multi-omics, and multi-organ frameworks in providing a comprehensive understanding of the human brain and its intricate connections to systemic health and disease.