The dawn of this advancement comes from the pioneering work led by Associate Professor Yao Zhigang at NUS’s Department of Statistics and Data Science. Traditional approaches to analyzing metabolomic profiles have long grappled with the high dimensionality and inherent noise of biomarker data. NMR, while rich in metabolic information, creates a multidimensional landscape where individual metabolites interact and fluctuate dynamically. The newly developed framework applies manifold fitting — a mathematical technique that models high-dimensional data onto low-dimensional, continuous surfaces called manifolds — to efficiently capture the intrinsic geometric structure underlying metabolomic variations.

At its core, this approach revolved around 251 distinct metabolic biomarkers measured in over 210,000 participants from the UK Biobank. By intelligently grouping these biomarkers into seven biologically coherent categories, reflecting known metabolic modules, the researchers could apply manifold fitting iteratively to each cluster. This strategy not only preserved the biological significance but also facilitated a smoother, noise-reduced representation of metabolic states. The result is a compelling geometric mapping where individual metabolic profiles are expressed as coordinates on a lower-dimensional manifold, revealing patterns and clusters invisible to conventional analytic tactics.

The power of this manifold approach lies in its ability to delineate subtle metabolic heterogeneity that aligns closely with clinically relevant health outcomes. Remarkably, in three of the seven biomarker groups, the researchers discovered that the manifolds naturally bifurcated the population into two distinct subgroups. Each subgroup exhibited differential disease risk associations, particularly for metabolic syndrome, cardiovascular disease, and autoimmune disorders. This refined stratification dramatically enhances the predictive value of metabolic profiling, potentially enabling earlier and more targeted interventions.

In a plenary lecture delivered at the 2025 International Congress of Chinese Mathematicians (ICCM), Associate Professor Yao emphasized the transformative potential of this methodology, stating, “By embedding high-dimensional biomarker data into interpretable, low-dimensional manifolds, we unlock new vistas for understanding metabolic diversity. This breakthrough opens pathways to more accurate risk prediction and personalized therapeutic strategies.” The machine learning-inspired framework thus bridges abstract mathematical theory with urgent biomedical challenges.

Beyond stratification, the manifold fitting method outperformed traditional statistical techniques in faithfully preserving biological signals while filtering out confounding noise. This fidelity is crucial, as metabolic biomarkers are not isolated measurements but part of an interconnected system influenced by genetics, lifestyle, and environmental factors. The enhanced interpretability and robustness of the method pave the way for integrating diverse data sources to decode the complex metabolic landscape.

Looking ahead, the research team envisions several exciting avenues to expand the scope and impact of their framework. One promising direction involves the integration of genomic data within the manifold-defined subpopulations. By performing genome-wide association studies (GWAS) tailored to each metabolically distinct cluster, they anticipate uncovering genetic variants that contribute uniquely to metabolic regulation and disease susceptibility. Such insights can illuminate the hereditary architecture behind metabolic heterogeneity, offering new targets for therapeutic intervention and risk assessment.

Another fertile area of investigation focuses on longitudinal dynamics. Metabolism is inherently fluid and responsive to temporal factors such as aging, diet, and disease progression. By applying manifold fitting to time-series metabolomic data, the team aims to map the trajectories individuals follow along metabolic manifolds over months or years. Tracking these transitions could reveal critical windows for disease onset or metabolic tipping points, facilitating proactive monitoring and intervention before irreversible damage occurs.

The versatility of manifold fitting also implies potential extensions into other omics layers, including proteomics and transcriptomics, further enriching the multidimensional characterizations of health and disease. As these data sources become more accessible, their integrative analysis through geometric frameworks promises even finer resolution in decoding biological complexity.

This trailblazing work, recently published in the prestigious Proceedings of the National Academy of Sciences (PNAS), signals a paradigm shift in metabolomic research. By reconciling the high dimensionality of NMR biomarker data with advanced geometric modeling, it lays a robust foundation for future explorations into metabolic disorders and personalized medicine. As Associate Professor Yao poignantly notes, “Our approach not only reveals current metabolic variation but also sets the stage to trace its genetic origins and chart its evolution in time — critical steps toward truly personalized healthcare.”

In an era dominated by big data and precision medicine, such innovative methodologies mark crucial progress in translating complex biological signals into actionable clinical insights. The manifold fitting framework offers a blueprint for tackling metabolic intricacies with unprecedented clarity, driving forward a future where tailored health interventions are informed by deep mathematical and biological understanding.

As the research community rallies around these promising techniques, the broader medical field stands poised to benefit from a new class of predictive tools capable of discerning individual risk profiles with greater nuance. Ultimately, this study exemplifies the profound impact interdisciplinary collaboration — between statisticians, mathematicians, biologists, and clinicians — can have in unlocking the secrets held within the human metabolome.

Subject of Research: Cells

Article Title: Manifold fitting reveals metabolomic heterogeneity and disease associations in UK Biobank populations

News Publication Date: 29 May 2025

Web References: https://doi.org/10.1073/pnas.2500001122

References: Proceedings of the National Academy of Sciences, 2025

Image Credits: National University of Singapore

Keywords: Applied mathematics, Mathematical analysis

Colon cancer remains one of the most prevalent and lethal malignancies globally, with survival heavily dependent on disease stage at diagnosis. While current prognostic models predominantly rely on pathological staging and demographic factors, they often lack sufficient sensitivity to anticipate patient outcomes. Addressing this gap, the recent study leverages advanced proteomic technologies to explore the plasma protein milieu years before clinical diagnosis, hypothesizing that early molecular alterations in circulating proteins could herald tumor behavior and patient prognosis.

Using plasma collected an average of nearly eight years before colon cancer diagnosis from participants in the extensive UK Biobank cohort, the research team applied Olink proteomics technology, a cutting-edge platform enabling high-throughput quantification of numerous proteins simultaneously with remarkable accuracy. This approach allowed the interrogation of protein landscapes long before tumor detection, offering unprecedented insight into the tumor microenvironment’s precancerous alterations.

The study delineates two distinct proteomic profiles corresponding to early and late stages of colon cancer, highlighting a temporal and biological complexity that challenges conventional paradigms. In early-stage cases, a 10-protein panel emerged, implicating biological processes such as extracellular matrix remodeling and immune evasion. These findings suggest that even before cancer is clinically evident, significant perturbations in the tissue scaffold and immune surveillance mechanisms are underway, potentially setting the stage for malignant transformation.

Specifically, the deregulation of innate immune activation pathways was prominent in the early-stage proteomic signature. This observation aligns with the growing understanding that cancer progression is not merely a result of tumor-intrinsic events but also reflects the dynamic interplay with the host immune system. The immune evasion tactics captured in the plasma proteome seem to foreshadow more aggressive disease courses, correlating with poorer survival post-diagnosis.

On the other hand, late-stage colon cancer exhibited a distinct 8-protein pre-diagnosis profile that intertwined pathological hallmarks of cell adhesion, angiogenesis, and pro-inflammatory responses. These processes are intimately linked