MSACL 2025 Abstract

Metabolic Profiling of Tissues for Biomarker Discovery in Breast Cancer

Valdis Gunnarsdottir Thormar (1), Kristrun Yr Holm (1), Yuchen Xiang (2), Zoltan Takats (2), Margret Thorsteinsdottir (1), Sigridur Klara Bodvarsdottir (1)
(1) University of Iceland, Reykjavik, Iceland, (2) Imperial College London, United Kingdom

Valdis Gunnarsdottir Thormar (Presenter) University of Iceland

Presenter Bio: Doctoral researcher in health sciences with the University of Iceland

Relevant Financial Disclosures (within past 24 months, reported on Apr 22, 2026) No relevant financial relationship(s) to disclose. Conflict mitigation NOT REQUIRED.

INTRODUCTION:
Breast cancer (BC) is the most prevalent cancer worldwide and the second leading cause of cancer-related deaths in women. Challenges relating to timely detection and the heterogeneous nature of BC complicate treatment strategies and influence patient outcomes. Conventional X-ray mammography screening, commonly used, exhibits limited sensitivity, particularly in young women with dense breast tissue, and lacks specificity for accurate BC diagnosis. Advanced diagnostic techniques that better characterize the disease enable more personalized treatment approaches, which improves prognosis.

OBJECTIVE(S):
This study aims to reveal associations between metabolites, lipids, and BC-related parameters, thereby facilitating the discovery of BC-specific biomarkers.

METHODS:
In this study, we use a metabolic fingerprinting method for analyzing fresh frozen (FF) tissue samples and archived formalin-fixed and paraffin-embedded (FFPE) tissue microarray (TMA) samples, on a well-defined Icelandic BC study cohort. Additionally, targeted and untargeted metabolomics analyses, covering both metabolites and lipids, have been performed on plasma samples from a subset of the same study cohort using LC-MS and NMR. The comprehensive metabolomics profiling of tissue was performed using desorption electrospray ionization mass spectrometry imagining (DESI-MSI) on TMAs from 222 BC patients and 30 normal tissues for diagnostic purposes.

Advanced machine learning techniques for predictive modeling will be applied to the data. By combining the analysis of metabolites and lipids with various BC-related parameters, we aim to explore underlying associations in the data. Both univariate and multivariate methods will be applied, as each provides distinct information about the data structure. Unsupervised algorithms like PCA, t-SNE, and UMAP will be used to visualize and elucidate data patterns, whilst supervised algorithms such as Gradient Descent, SVM, and Random Forest, including ensemble learning techniques, will be deployed to build robust classifiers. Feature selection methods will be implemented to help identify biomarkers and regulate model complexity. To ensure model reliability and generalizability, rigorous cross-validation and data partitioning into training, testing, and validation sets are conducted. Partitioning and cross-validation allow for the robust evaluation of how well the models perform on new, unseen data.

RESULTS:
DESI-MSI has shown the ability to distinguish between tumor and normal breast tissues based on their metabolomics profiles. Additionally, the results showed that despite formalin fixation and paraffin embedding on the tissue, enough information could be obtained to characterize the tissue. To confirm and expand upon these findings, the DESI-MSI data is being pre-processed again according to a new pipeline, effectively increasing the number of detectable metabolites in FFPE BC tissue and lipids in FF BC tissues.

CONCLUSION:
In this ongoing research, we investigate the correlations between BC subtypes, relevant patient characteristics, and their metabolomic profiles, aiming at a holistic view of the biomolecular changes associated with BC. Ultimately, to identify novel biomarkers.