Metabolic detection of early-stage breast cancer through absolute quantitative metabolomics and machine learning

To achieve the best outcomes, breast cancer necessitates robust strategies for early detection. However, reliable blood-based tests for identifying early-stage disease remains elusive. Here we have employed plasma metabolomics and machine learning techniques to establish a non-invasive metabolic approach for early detection of breast cancer. Utilizing support vector machine algorithms and mass spectrometry, four characteristic metabolites (inosine, uridine, phenylalanine and threonine) were identified and optimized as detection features. A liquid chromatography–mass spectrometry (LC/MS)–based targeted assay was then developed for concurrent absolute quantitation of these four metabolites. The study included 1111 participants, and a predictive model based on support vector machine achieved high accuracy of 94.09%, 90.94%, and 89.29%, with area under the receiver-operating characteristic curves of 0.978, 0.950, and 0.926 for the training, test, and independent validation cohorts, respectively. Notable alterations in the four metabolic pathways and key metabolic genes regulating levels of the selected metabolites in breast tumor cells were discovered through single-cell RNA sequencing analysis. This study introduces a promising non-invasive metabolic method for early breast cancer detection and offers insights into the molecular mechanisms underpinning metabolic dysregulation in breast cancer.