Quantifying Individual Health Status from Multi-omics Data by Health State Manifold

X. Y. Zhang, Chengming Zhang, Yanpu Wu, Xufeng Lin, Xiaoping Liu, Luonan ChenPhenomics 2025 · DOI

TL;DR

This paper introduces a novel "Health State Manifold" (HSM) method to quantify an individual's health status by integrating multi-omics data. HSM decomposes health into two key components: phenotypic state (where you are on the health-to-disease spectrum) and homeostatic resilience (your ability to maintain that state), offering a more nuanced view than traditional methods. This approach has shown promise in distinguishing subtle health differences and identifying critical transition points in various disease models.

What Problem Does This Paper Address?

The paper tackles the challenge of accurately quantifying an individual's health status, especially in the early stages of disease or during recovery. Current approaches often rely on static biomarkers or classify individuals into broad categories, which can miss subtle changes or individuals with similar phenotypes but different risks of deterioration. The authors argue that health is a dynamic process, not just the absence of disease, and that understanding an individual's ability to withstand perturbations (homeostatic resilience) is as crucial as knowing their current state (phenotypic state). This is particularly important for personalized medicine and preventive strategies, where early detection and intervention are key.

Methods

The core of the HSM method lies in its integration of two key concepts:

  • Dynamic Network Biomarker (DNB) theory: This framework is used to understand how biological systems change over time and identify critical "tipping points" or pre-disease states.
  • Diffusion Map Theory: This technique is employed to reduce high-dimensional omics data into a lower-dimensional "health process manifold."

Specifically, the HSM method quantifies health using two scores:

  1. V score (Phenotypic State/Potential): This score represents an individual's position on the health process manifold. It's calculated using a Linear Discriminant Analysis (LDA) model combined with Laplacian Eigenmaps (LDA-LE). This approach aims to maximize the separation between "healthy" (background) and "diseased" (target) groups while preserving the local structure of the data, effectively mapping individuals onto a spectrum from health to disease.
  2. U score (Homeostatic Resilience/Potential): This score quantifies an individual's ability to maintain their current health state. It's derived from the minimum principal curvature of a "homeostatic potential landscape," which is constructed from time-course omics data using DNB principles. A higher U score indicates greater stability and resilience.

The paper also details a tenfold cross-validation procedure to ensure the robustness of the V score calculation by identifying stable features.

Datasets

The HSM method was validated using several publicly available datasets:

  • Rats Diabetes Dataset (GSE13270): Time-series gene expression data from rats modeling type 2 diabetes progression and insulin resistance.
  • Human Roux-en-Y Gastric Bypass (RYGB) Dataset: Time-series metabolite data from individuals undergoing RYGB surgery, tracking recovery.
  • Human Hepatocellular Carcinoma (HCC) Dataset (GSE6764): Gene expression data representing different stages of liver cancer development.
  • Mice Diabetes Dataset (GSE15150): Time-series gene expression data from non-obese diabetic mice.
  • Mice Lung Adenocarcinoma (LUAD) Dataset (GSE102707): Gene expression data from mice exposed to carcinogens, modeling lung cancer development.

These datasets cover various disease contexts and biological levels (gene expression, metabolites), demonstrating the broad applicability of the HSM approach.

Key Insights and Ideas

  • Health is a dynamic spectrum, not a binary state: The HSM method moves beyond simple classification to represent health as a continuous process with distinct stages and varying stability.
  • Homeostatic resilience is a critical, often overlooked, health indicator: The U score highlights that individuals with similar disease states can have vastly different risks of deterioration based on their resilience.
  • Tipping points and pre-disease states can be identified: The U score effectively pinpoints moments of low homeostatic potential, which correspond to critical transition points or pre-disease states, offering a window for early intervention.
  • Multi-omics integration is key: Combining phenotypic state (V score) and homeostatic resilience (U score) provides a more comprehensive and accurate picture of an individual's health than either metric alone.
  • Potential for identifying disease-related genes: The study suggests that genes contributing to low homeostatic potential might be key players in disease transition and progression.

Key Results

The HSM method was successfully applied to all tested datasets, demonstrating its effectiveness in:

  • Distinguishing individuals with similar phenotypes: In the rat diabetes dataset, individuals with similar V scores (phenotypic state) could be differentiated by their U scores (homeostatic resilience), revealing different risk levels.
  • Quantifying disease progression and recovery: The V scores clearly mapped the progression of diabetes in rats, the recovery process after RYGB surgery in humans, and the stages of HCC and LUAD development in mice.
  • Identifying critical transition points: In several datasets, low U scores coincided with known critical stages of disease development or instability (e.g., 8 weeks in GK rats for T2DM, 1 week post-RYGB for recovery).
  • Validating findings with existing literature: The identified critical points and disease associations were consistent with previous biological studies.

Limitations and Future Work

The authors acknowledge several limitations and suggest future directions:

  • Systemic Complexity: The human body is a complex system with interacting tissues and pathways. The current approach might benefit from more targeted feature selection for specific diseases.
  • Tissue Specificity: For certain diseases, focusing on omics data from the affected tissue might yield more precise results.
  • Homeostatic Potential Calculation: The current U score calculation requires multi-sample data from a single state. Developing a method to calculate homeostatic potential from single-sample data is a key area for future work.
  • Experimental Validation: While the method shows promise, further joint experimental and theoretical studies are needed to fully understand the essential state transition laws of individual health.
  • Potential Disease Gene Identification: The study provides preliminary insights into potential disease genes, but further research is needed to validate these findings and explore their functional roles.

Summary updated

Abstract

The online version contains supplementary material available at 10.1007/s43657-024-00188-4.
BibTeX
@inproceedings{zhang2025quantifying,
  author = {X. Y. Zhang and Chengming Zhang and Yanpu Wu and Xufeng Lin and Xiaoping Liu and Luonan Chen},
  title = {{Quantifying Individual Health Status from Multi-omics Data by Health State Manifold}},
  booktitle = {Phenomics},
  year = {2025},
  doi = {10.1007/s43657-024-00188-4},
}
Figure