Multi-organ AI Endophenotypes Chart the Heterogeneity of Pan-disease in the Brain, Eye, and Heart

T. Consortium, Aleix Boquet-Pujadas, Filippos Anagnostakis, Zhijian Yang, Ye Tian, Michael R. Duggan, Güray Erus, Dhivya Srinivasan, Cassandra M Joynes, Wenjia Bai, Praveen J. Patel, Keenan A. Walker, Andrew Zalesky, Christos DavatzikosmedRxiv 2025 · DOI

Abstract

Disease heterogeneity and commonality pose significant challenges to precision medicine, as traditional approaches frequently focus on single disease entities and overlook shared mechanisms across conditions. Inspired by pan-cancer and multi-organ research, we introduce the concept of "pan-disease" to investigate the heterogeneity and shared etiology in brain, eye, and heart diseases. Leveraging individual-level data from 129,340 participants, as well as summary-level data from the MULTI consortium, we applied a weakly-supervised deep learning model (Surreal-GAN) to multi-organ imaging, genetic, proteomic, and RNA-seq data, identifying 11 AI-derived biomarkers - called Multi-organ AI Endophenotypes (MAEs) - for the brain (Brain 1-6), eye (Eye 1-3), and heart (Heart 1-2), respectively. We found Brain 3 to be a risk factor for Alzheimer's disease (AD) progression and mortality, whereas Brain 5 was protective against AD progression. Crucially, in data from an anti-amyloid AD drug (solanezumab), heterogeneity in cognitive decline trajectories was observed across treatment groups. At week 240, patients with lower brain 1-3 expression had slower cognitive decline, whereas patients with higher expression had faster cognitive decline. A multi-layer causal pathway pinpointed Brain 1 as a mediational endophenotype linking the FLRT2 protein to migraine, exemplifying novel therapeutic targets and pathways. Additionally, genes associated with Eye 1 and Eye 3 were enriched in cancer drug-related gene sets with causal links to specific cancer types and proteins. Finally, Heart 1 and Heart 2 had the highest mortality risk and unique medication history profiles, with Heart 1 showing favorable responses to antihypertensive medications and Heart 2 to digoxin treatment. The 11 MAEs provide novel AI dimensional representations for precision medicine and highlight the potential of AI-driven patient stratification for disease risk monitoring, clinical trials, and drug discovery.

Introduction

Precision medicine often operates within organ-specific silos, treating conditions like Alzheimer’s disease, glaucoma, and heart failure as isolated entities. However, biological processes frequently transcend these boundaries. This study introduces a "pan-disease" framework to map shared pathological patterns across the brain-eye-heart axis. By moving beyond binary ICD-10 diagnoses, the researchers aim to identify "intermediate phenotypes"—measurable biological traits that capture the spectrum of disease heterogeneity and commonality.

Methodology: The MAE Framework

The study utilized a weakly-supervised deep learning architecture, Surreal-GAN, to identify distinct dimensions of structural change. Unlike traditional clustering, Surreal-GAN learns a "1-to-k" mapping by comparing a healthy reference population to a "pan-disease" population (individuals with various diagnoses). This allows for the decomposition of complex disease presentations into $k$ continuous biomarkers, termed Multi-organ AI Endophenotypes (MAEs).

Overview of the pan-disease framework, including Surreal-GAN architecture, multi-organ data integration, and clinical outcome prediction.

The researchers integrated these imaging-derived MAEs with a comprehensive multi-omics stack, including genetics (GWAS), proteomics (linking to ~3,000 plasma proteins), and causal inference via Mendelian Randomization. Data was sourced from the MULTI consortium, encompassing 129,340 participants from the UK Biobank (UKBB) for discovery, with external validation in cohorts such as BLSA, A4, and ADNI.

Results: Identifying 11 Multi-organ AI Endophenotypes

The analysis identified a total of 11 MAEs: six for the brain, three for the eye, and two for the heart. The selection of the optimal number of dimensions ($k$) for each organ was based on the stability and correlation of R-indices across different hyperparameter settings.

Selection of optimal k for brain (k=6), eye (k=3), and heart (k=2) based on R-indices correlation across varying gamma and lambda parameters.

Clinical and Biological Significance

  • Alzheimer’s Disease (AD) and Mortality: Brain 3 was identified as a significant risk factor for AD progression and mortality. Conversely, Brain 5 appeared to be "protective," with higher expression correlating with slower cognitive decline.
  • Drug Trial Stratification: Re-analysis of the solanezumab trial data revealed that the drug's efficacy was highly dependent on the patient's MAE profile. Patients with lower Brain 1-3 expression showed significantly slower cognitive decline at week 240, suggesting that MAEs could be used to identify responders in "failed" clinical trials.
  • Causal Pathways: A vertical causal pathway was established linking the FLRT2 protein to temporal/frontal lobe atrophy (Brain 1), which in turn increases migraine risk.
  • Systemic Connections: Eye MAEs (1 and 3) showed genetic enrichment in cancer-related pathways, reinforcing the eye's role as a window into systemic health. Heart MAEs were the strongest predictors of overall mortality and showed distinct medication response profiles (e.g., Heart 1 and antihypertensives).

Discussion

The 11 MAEs provide a novel dimensional representation of disease that outperforms traditional categorical diagnoses in predicting clinical outcomes and drug responses. This pan-disease approach highlights the interconnectedness of organ systems and the utility of AI in uncovering hidden biological commonalities.

While promising, the study notes limitations including a primary focus on European-ancestry populations and the need for further external validation of heart and eye MAEs. Future work will focus on expanding these models to more diverse populations and transitioning these research tools into clinically actionable tests.

Source: Multi-organ AI Endophenotypes Chart the Heterogeneity of Pan-disease in the Brain, Eye, and Heart. (2025). medRxiv preprint. https://doi.org/10.1101/2025.08.09.25333350

Summary updated

Abstract

Disease heterogeneity and commonality pose significant challenges to precision medicine, as traditional approaches frequently focus on single disease entities and overlook shared mechanisms across conditions<sup>1</sup>. Inspired by pan-cancer<sup>2</sup> and multi-organ research<sup>3</sup>, we introduce the concept of "pan-disease" to investigate the heterogeneity and shared etiology in brain, eye, and heart diseases. Leveraging individual-level data from 129,340 participants, as well as summary-level data from the MULTI consortium, we applied a weakly-supervised deep learning model (Surreal-GAN<sup>4,5</sup>) to multi-organ imaging, genetic, proteomic, and RNA-seq data, identifying 11 AI-derived biomarkers - called Multi-organ AI Endophenotypes (MAEs) - for the brain (Brain 1-6), eye (Eye 1-3), and heart (Heart 1-2), respectively. We found Brain 3 to be a risk factor for Alzheimer's disease (AD) progression and mortality, whereas Brain 5 was protective against AD progression. Crucially, in data from an anti-amyloid AD drug (solanezumab<sup>6</sup>), heterogeneity in cognitive decline trajectories was observed across treatment groups. At week 240, patients with lower brain 1-3 expression had slower cognitive decline, whereas patients with higher expression had faster cognitive decline. A multi-layer causal pathway pinpointed Brain 1 as a mediational endophenotype<sup>7</sup> linking the FLRT2 protein to migraine, exemplifying novel therapeutic targets and pathways. Additionally, genes associated with Eye 1 and Eye 3 were enriched in cancer drug-related gene sets with causal links to specific cancer types and proteins. Finally, Heart 1 and Heart 2 had the highest mortality risk and unique medication history profiles, with Heart 1 showing favorable responses to antihypertensive medications and Heart 2 to digoxin treatment. The 11 MAEs provide novel AI dimensional representations for precision medicine and highlight the potential of AI-driven patient stratification for disease risk monitoring, clinical trials, and drug discovery.
BibTeX
@misc{consortium2025multi,
  author = {T. Consortium and Aleix Boquet-Pujadas and Filippos Anagnostakis and Zhijian Yang and Ye Tian and Michael R. Duggan and Güray Erus and Dhivya Srinivasan and Cassandra M Joynes and Wenjia Bai and Praveen J. Patel and Keenan A. Walker and Andrew Zalesky and Christos Davatzikos},
  title = {{Multi-organ AI Endophenotypes Chart the Heterogeneity of Pan-disease in the Brain, Eye, and Heart}},
  year = {2025},
  doi = {10.1101/2025.08.09.25333350},
  howpublished = {medRxiv},
}
Figure