Diagnosing lung cancer through metabolic fingerprint based on machine learning
10.3760/cma.j.cn114452-20211109-00696
- VernacularTitle:血清代谢组指纹联合机器学习诊断肺癌
- Author:
Yuxin ZHANG
1
;
Chengwen HE
;
Lin HUANG
;
Kun QIAN
;
Wei CHEN
;
Yin JIA
;
Jingjing HU
;
Qin WEI
;
Xiping WANG
;
Shanrong LIU
Author Information
1. 上海交通大学生物医学工程学院和医疗机器人研究所，上海200030
- Keywords:
Mass spectrometry;
Lung cancer;
Metabolic fingerprinting;
Machine learning;
Experimental diagnosis
- From:
Chinese Journal of Laboratory Medicine
2022;45(3):226-233
- CountryChina
- Language:Chinese
-
Abstract:
Objective:To screen out the differentially regulated metabolites by the analysis of serum metabolic fingerprints, and to provide potential biomarkers for diagnosis of lung cancer.Methods:A total of 228 subjects were enrolled in Changhai Hospital from January 27, 2021 to June 4, 2021, including 97 newly diagnosed lung cancer patients and 131 healthy individuals. Serum samples were collected from the enrolled cohort according to a standard procedure, and the enrolled cohort was divided into a training set and a completely independent validation set by stratified random sampling. The metabolic fingerprints of serum samples were collected by previously developed nano-assisted laser desorption/ionization mass spectrometry (nano-LDI MS). After age and gender matching of the training set, a diagnostic model based on serum metabolic fingerprints was established by machine learning algorithm, and the classification performance of the model was evaluated by receiver operating characteristic (ROC) curve.Results:Serum metabolic fingerprint for each sample was obtained in 1 minute using a novel nano-LDI MS, with consumption of only 1 μl original serum sample. For the training set, the area under ROC curve (AUC) of the constructed classifier for diagnosis of lung cancer was 0.92 (95% CI 0.87-0.97), with a sensitivity of 89% and specificity of 89%. For the independent validation set, the AUC reached 0.96 (95% CI 0.90-1.00) with a sensitivity of 91% and specificity of 94%, which showed no significant decrease compared to training set. We also identified a biomarker panel of 5 metabolites, demonstrating a unique metabolic fingerprint of lung cancer patients. Conclusion:Serum metabolic fingerprints and machine learning were combined to establish a diagnostic model, which can be used to distinguish between lung cancer patients and healthy controls. This work sheds lights on the rapid metabolic analysis for clinical application towards in vitro diagnosis.
