Integration of metabolomics and machine learning revealed tryptophan metabolites are sensitive biomarkers of pemetrexed efficacy in non‐small cell lung cancer

Abstract

Abstract Background Anti‐folate drug pemetrexed is a vital chemotherapy medication for non‐small cell lung cancer (NSCLC). Its response varies widely and often develops resistance to the treatment. Therefore, it is urgent to identify biomarkers and establish models for drug efficacy evaluation and prediction for rational drug use. Methods A total of 360 subjects were screened and 323 subjects were recruited. Using metabolomics in combination with machine learning methods, we are trying to select potential biomarkers to diagnose NSCLC and evaluate the efficacy of pemetrexed in treating NSCLC. Furtherly, we measured the concentration of eight metabolites in the tryptophan metabolism pathway in the validation set containing 201 subjects using a targeted metabolomics method with UPLC‐MS/MS. Results In the discovery set containing 122 subjects, the metabolic profile of healthy controls (H), newly diagnosed NSCLC patients (ND), patients who responded well to pemetrexed treatment (S) and pemetrexed‐resistant patients (R) differed significantly on the PLS‐DA scores plot. Pathway analysis showed that glycine, serine and threonine metabolism occurred in every two group comparisons. TCA cycle, pyruvate metabolism and glycerolipid metabolism are the most significantly changed pathways between ND and H group, pyruvate metabolism was the most altered pathway between S and ND group, and tryptophan metabolism was the most changed pathway between S and R group. We found Random forest method had the maximum area under the curve (AUC) and can be easily interpreted. The AUC is 0.981 for diagnosing patients with NSCLC and 0.954 for evaluating pemetrexed efficiency. Conclusion We compared eight mathematical models to evaluate pemetrexed efficiency for treating NSCLC. The Random forest model established with metabolic markers tryptophan, kynurenine and xanthurenic acidcan accurately diagnose NSCLC and evaluate the response of pemetrexed.

Keywords

• drug resistance
• machine learning
• NSCLC
• pemetrexed
• pharmacometabolomics