Machine learning-assisted quantitative metabolomics of West African patients with advanced breast cancer.

In this study, we conducted a targeted quantitative metabolomic analysis of 630 metabolites in the plasma of 78 West African patients at the time of breast cancer diagnosis and prior to any treatment. Most of these patients were at an advanced stage of the disease. The data were compared with those of 79 healthy controls using a combination of several machine learning approaches and statistical analyses. The predictive models obtained with the machine learning algorithms were comparable, with the best AUC of 0.878 obtained with ridge logistic regression using Boruta feature selection. The most consistently identified discriminating metabolites across univariate analyses with Benjamini-Hochberg correction, OPLS-DA analyses, and the best machine learning approach were thirteen, out of a total of 63 discriminating metabolites identified cumulatively by the three approaches. This signature highlights several key biological processes, including oxidative stress, disrupted neurotransmitter profiles, altered nitric oxide and xanthine oxidase metabolism, and impaired energy metabolism. The involvement of new metabolites significantly deregulated in breast cancer, such as asymmetric dimethylarginine and hexosylceramides, have also been identified. The identified metabolomic signature provides a comprehensive and global view of the blood biochemical phenotype associated with advanced breast cancer at the time of diagnosis.