Early detection of renal cell carcinoma through machine learning analysis of metabolomic signature

Renal cell carcinoma is a common, heterogeneous cancer with variable prognosis. For effective treatment and thus improvement of patient outcome, early and accurate diagnosis of RCC is imperative. The present study evaluates the potential of metabolomics, the study of all small molecules in biological material, as a diagnostic tool in RCC. An XGBoost machine learning model was developed on 9401 metabolomic features to differentiate healthy individuals from those with RCC and also differentiate patients with varying stages of RCC. Control data were obtained from the NIH Common Fund's National Metabolomics Data Repository (PR001932). RCC metabolomic data was sourced from the supplementary material of Jing et al. (2019). Feature selection using the Boruta algorithm identified 14 key metabolites significantly associated with RCC. The performance of the XGBoost model, after training, on a held-out test set was 88% accuracy, 96% precision, 100% recall, and an F1-score of 98%, demonstrating the potential of metabolomic profiling combined with machine learning for non-invasive RCC diagnosis. This approach holds promise for improving early detection and personalized management of RCC.