Combination of Plasma-Based Metabolomics and Machine Learning Algorithm Provides a Novel Diagnostic Strategy for Malignant Mesothelioma

Our plain-English summary of Combination of Plasma-Based Metabolomics and Machine Learning Algorithm Provides a Novel Diagnostic Strategy for Malignant Mesothelioma, by N. Li, C. Yang, S. Zhou, et al (Diagnostics, 2021).

The study

• Authors: N. Li, C. Yang, S. Zhou, S. Song, Y. Jin, D. Wang, J. Liu, Y. Gao, H. Yang, W. Mao, Z. Chen
• Journal: Diagnostics
• Published: 2021;11(7)
• Cited by: 7 articles, as counted by NIH iCite
• Read it: PubMed · Publisher · Free full text

Abstract

Background: Malignant mesothelioma (MM) is an aggressive and incurable carcinoma that is primarily caused by asbestos exposure. However, the current diagnostic tool for MM is still under-developed. Therefore, the aim of this study is to explore the diagnostic significance of a strategy that combined plasma-based metabolomics with machine learning algorithms for MM.

Methods: Plasma samples collected from 25 MM patients and 32 healthy controls (HCs) were randomly divided into train set and test set, after which analyzation was performed by liquid chromatography-mass spectrometry-based metabolomics. Differential metabolites were screened out from the samples of the train set. Subsequently, metabolite-based diagnostic models, including receiver operating characteristic (ROC) curves and Random Forest model (RF), were established, and their prediction accuracies were calculated for the test set samples.

Results: Twenty differential plasma metabolites were annotated in the train set; 10 of these metabolites were validated in the test set. The seven most prevalent diagnostic metabolites were taurocholic acid), 0.7142 (uracil), 0.7142 (biliverdin), 0.8571 (histidine), 0.5000 (tauroursodeoxycholic acid), 0.8571 (pyrroline hydroxycarboxylic acid), and 0.7857 (phenylalanine). Furthermore, RF based on 20 annotated metabolites showed a prediction accuracy of 0.9286, and its optimized version achieved 1.0000 in the test set. Moreover, the comparison between the samples of peritoneal MM (n = 8) and pleural MM (n = 17) illustrated a significant increase in levels of taurocholic acid and tauroursodeoxycholic acid, as well as an evident decrease in biliverdin.

Conclusions: Our results revealed the potential diagnostic value of plasma-based metabolomics combined with machine learning for MM. Further research with large sample size is worthy conducting. Moreover, our data demonstrated dysregulated metabolism pathways in MM, which aids in better understanding of molecular mechanisms related to the initiation and development of MM.

In plain English

This study from Zhejiang Cancer Hospital, China, tested whether plasma metabolite profiles analysed by machine learning could distinguish malignant mesothelioma patients from healthy people. It included 25 patients whose plasma samples were collected between March 2016 and March 2020 (17 pleural, 8 peritoneal; 11 reported asbestos exposure, 10 did not, 4 unknown) and 32 age- and sex-matched healthy controls. Participants were randomly split into a training group of 43 and a test group of 14 (6 patients, 8 controls). In the training group, 20 metabolites differed between patients and controls, and 10 of these were confirmed in the test group. Single metabolites classified test-group samples with accuracies from 0.5000 to 0.8571. A Random Forest model using all 20 metabolites had a test-group accuracy of 0.9286, and a version using the top 9 reached 1.0000. Two bile acids, taurocholic acid and tauroursodeoxycholic acid, were higher in the 8 peritoneal than the 17 pleural patients. The authors concluded the approach showed potential diagnostic value. Because the sample was small, they called the study a pilot. They noted it lacked separate comparison groups of asbestos-exposed individuals and of patients with cancers easily mistaken for mesothelioma.

We wrote this plain-English summary from the full article with the help of AI, and a second AI model checked each sentence against the article. It is not the authors' wording; the abstract and the article are their own account.

What the study did and found

• Design: diagnostic case-control study using untargeted blood plasma metabolomics (LC-MS) combined with single-metabolite ROC models and a Random Forest machine learning model, with a random split into training and test sets
• Who was studied: 25 patients with histopathologically confirmed malignant mesothelioma (17 pleural, 8 peritoneal), none previously treated, at Zhejiang Cancer Hospital, China. Their plasma samples were collected between March 2016 and March 2020. There were also 32 age- and sex-matched healthy controls. Participants were split into a training set of 43 (19 patients, 24 controls) and a test set of 14 (6 patients, 8 controls).
• Exposure: Asbestos exposure was not the subject of the study. It was recorded for the patients only: 11 reported exposure, 10 reported none, and 4 were unknown. For healthy controls, exposure was listed as not available. No fiber types or jobs were given.
• Main results:
  • In the training set, 20 differential plasma metabolites were annotated (8 higher and 12 lower in patients). In the test set, 17 showed the same direction of change, and 10 of those were statistically significant.
  • Single-metabolite models in the test set had lower accuracy for some metabolites: taurocholic acid 0.6429, uracil 0.7143, biliverdin 0.7143 and tauroursodeoxycholic acid 0.5000.
  • Histidine and pyrroline hydroxycarboxylic acid each had a test-set accuracy of 0.8571 as single-metabolite models.
  • The Random Forest model using all 20 metabolites had an accuracy of 1.0000 in the training set and 0.9286 in the test set.
  • A Random Forest model using the top 9 ranked metabolites reached an accuracy of 1.0000 in the test set.
  • Comparing peritoneal (n = 8) with pleural (n = 17) mesothelioma patients, the results section reported tauroursodeoxycholic acid, taurocholic acid (printed as 'tauroucholic acid') and biliverdin as significantly higher in peritoneal patients. The article's own abstract instead describes biliverdin as decreased.
• The authors' conclusion: The authors concluded that plasma-based metabolomics combined with machine learning showed potential diagnostic value for malignant mesothelioma. They said further research with a large sample size is needed.
• Limitations the authors state:
  • The sample size was small for machine learning, and the authors described the study as possibly only a pilot study.
  • The model compared only two groups (patients and healthy controls). The authors said future work should include asbestos-exposed individuals and cancers easily misdiagnosed as mesothelioma, such as lung or ovarian cancer.
  • Few metabolites could be annotated because MS2 spectra were lacking, so many potentially useful features were not identified.
  • The authors attributed inconsistent bile acid results between the training and test sets to uneven allocation of peritoneal patients: only one was in the test set, compared with 7 in the training set.

Topics

Medical terms in this study, explained in our Lexicon: Malignant Mesothelioma, Carcinoma, Asbestos Exposure.

Related studies

• The use of a next-generation sequencing-derived machine-learning risk-prediction model (OncoCast-MPM) for malignant pleural mesothelioma: a retrospective study — M.G. Zauderer, A. Martin, et al. (2021)
• Prediction of Radiation Pneumonitis With Machine Learning in Stage III Lung Cancer: A Pilot Study — M. Yakar, D. Etiz, et al. (2021)
• Predicting cancer using supervised machine learning: Mesothelioma — A. Choudhury (2021)
• Differential diagnosis of pleural mesothelioma using Logic Learning Machine — S. Parodi, R. Filiberti, et al. (2015)
• Towards a potential pan-cancer prognostic signature for gene expression based on probesets and ensemble machine learning — D. Chicco, A. Alameer, et al. (2022)
• GC-MS-based untargeted metabolic profiling of malignant mesothelioma plasma — D. Wang, J. Zhu, et al. (2023)

Browse more research in Diagnostics, or return to medical research.