Trial Report Prediction of myalgic chronic fatigue syndrome disorder with machine learning approach, 2024, Yagin & Georgian

Discussion in 'ME/CFS research' started by Dolphin, Jul 6, 2024.

  1. Dolphin

    Dolphin Senior Member (Voting Rights)

    Messages:
    5,861
    Free fulltext:
    https://e-jespar.com/index.php/jespar/article/view/25

    Prediction of myalgic chronic fatigue syndrome disorder with machine learning approach

    Authors
    • Fatma Hilal Yagin
    • Badicu Georgian

    • Department of Physical Education and Special Motricity, Transilvania University of Brasov, 500068 Brasov, Romania
    DOI:
    https://doi.org/10.5281/zenodo.12601089

    Keywords:

    Metabolomics, machine learning, random forest, gaussian naive bayes

    Abstract


    Myalgic encephalomyelitis/chronic fatigue syndrome (ME/CFS) is a complex disorder characterized by unexplained fatigue, post-exertional malaise, unrefreshing sleep, and cognitive impairment or orthostatic intolerance.

    Due to the absence of a recognized laboratory diagnostic test, diagnosis relies on patient history and physical examination.

    This study aimed to identify significant metabolomic markers and employ machine learning techniques for the classification of ME/CFS.

    Utilizing open-access metabolomics data from 26 ME/CFS patients and 26 controls, we implemented a comprehensive data preprocessing and modeling framework.

    Feature selection was performed using Random Forest, and data normalization was achieved through standardization.

    A Gaussian Naive Bayes model was trained and validated using 5-fold cross-validation.

    The model exhibited an accuracy of 0.786, sensitivity of 0.952, specificity of 0.619, and an F1 score of 0.816.

    These results indicate a high efficacy in identifying positive cases of ME/CFS.


    Yagin , F. H. ., & Georgian, B. (2024). Prediction of myalgic chronic fatigue syndrome disorder with machine learning approach. Journal of Exercise Science & Physical Activity Reviews, 2(1), 97–103. https://doi.org/10.5281/zenodo.12601089
     
    Sean, Kitty, Hutan and 2 others like this.
  2. Hutan

    Hutan Moderator Staff Member

    Messages:
    29,562
    Location:
    Aotearoa New Zealand
    It's great to see interest in ME/CFS in Romania.

    But this paper has some odd features. It is published in the Journal of Exercise Science and Physical Activity Reviews, but it is not a review and it is not about exercise science.

    It gives the appearance of a preprint - the highlights below are in the paper (they are not my additions). It is not clear what study was the source of the data, as shown by the red highlight.

    Screen Shot 2024-07-07 at 5.42.11 am.png

    The DOI doesn't work, at least not yet.
     
    MeSci, Sean, alktipping and 8 others like this.
  3. forestglip

    forestglip Senior Member (Voting Rights)

    Messages:
    987
    Same author, and also has 26 ME/CFS and 26 controls, all female: An Explainable Artificial Intelligence Model Proposed for the Prediction of Myalgic Encephalomyelitis/Chronic Fatigue Syndrome and the Identification of Distinctive Metabolites, 2023, Yagin et al (Thread)

    Source given in that paper for the data: Comprehensive Circulatory Metabolomics in ME/CFS Reveals Disrupted Metabolism of Acyl Lipids and Steroids, 2020, Germain et al (Thread)
     
    Last edited: Jul 6, 2024
    MeSci, Sean, Hutan and 6 others like this.
  4. Sid

    Sid Senior Member (Voting Rights)

    Messages:
    1,190
    Machine learning with n=26 will result in an overfitted model that likely won’t replicate elsewhere.
     
    alktipping, rvallee, Dolphin and 9 others like this.
  5. Sean

    Sean Moderator Staff Member

    Messages:
    8,154
    Location:
    Australia
    Yep. Needs at least 2-3 more orders of magnitude in the sample size.
     

Share This Page