¹ H NMR-Based Plasma Metabolomics and Machine Learning Reveal Distinct Metabolic Signatures of Breast Cancer in Nigerian Women

Abstract

Untargeted 1H NMR metabolomics offers a noninvasive means to identify biomarkers in breast cancer (BC) patients; however, metabolic signatures specific to Nigerian women remain poorly understood. This study aimed to identify plasma metabolomic and lipidomic biomarkers associated with BC in Nigerian patients, evaluate their diagnostic performance using machine learning (ML), and identify dysregulated metabolic pathways. This case–control study recruited 100 BC patients and 100 healthy controls from 4 Nigerian teaching hospitals. Plasma metabolites and lipids were profiled using 1H NMR spectroscopy and the Liposcale test. Multivariate and ML analyses revealed a clear distinction between BC and controls (PLS-DA accuracy: 92.4–94.4%). Twenty-four metabolites were significantly altered (FDR < 0.05), with decreased glycine and glutamine, and increased GlycA, GlycB, glucose, and ketone bodies. Lipoprotein profiling showed reduced small HDL, LDL, and large VLDL particles, alongside with increased HDL diameter. The random forest model achieved the best classification performance (AUC = 0.985) and identified 23 key biomarkers. Pathway analysis revealed 29 enriched metabolic pathways, including glyoxylate and dicarboxylate metabolism. Overall, these findings highlight distinct metabolic alterations in Nigerian BC patients and demonstrate the potential of combining NMR-based metabolomics with ML for population-specific, noninvasive BC diagnostics.