Diagnosis of renal cell carcinoma (RCC) at an early stage is challenging, but it provides the best chance for cure. We aimed to develop a predictive diagnostic method for early-stage RCC based on a biomarker cluster using nuclear magnetic resonance (NMR)-based serum metabolomics and self-organizing maps (SOMs). We trained and validated the SOM model using serum metabolome data from 104 participants, including healthy individuals and early-stage RCC patients. To assess the predictive capability of the model, we analyzed an independent cohort of 22 subjects. We then used our method to evaluate changes in the metabolic patterns of 23 RCC patients before and after nephrectomy. A biomarker cluster of 7 metabolites (alanine, creatine, choline, isoleucine, lactate, leucine, and valine) was identified for the early diagnosis of RCC. The trained SOM model using a biomarker cluster was able to classify 22 test subjects into the appropriate categories. Following nephrectomy, all RCC patients were classified as healthy, which was indicative of metabolic recovery. But using a diagnostic criterion of 0.80, only 3 of the 23 subjects could not be confidently assessed as metabolically recovered after nephrectomy. We successfully followed-up 17 RCC patients for 8 years post-nephrectomy. Eleven of these patients who diagnosed as metabolic recovery remained healthy after 8 years. Our data suggest that a SOM model using a biomarker cluster from serum metabolome can accurately predict early RCC diagnosis and can be used to evaluate postoperative metabolic recovery.
Keywords: artificial intelligence; early diagnosis; metabolic recovery; metabolome; precision medicine.