PSR-MAPMS: A new approach for the interpretable prediction of myelin autoantigenic peptides in multiple sclerosis using multi-source propensity scores
2
Issued Date
2025-08-01
Resource Type
ISSN
09618368
eISSN
1469896X
Scopus ID
2-s2.0-105010931858
Journal Title
Protein Science
Volume
34
Issue
8
Rights Holder(s)
SCOPUS
Bibliographic Citation
Protein Science Vol.34 No.8 (2025)
Suggested Citation
Charoenkwan P., Schaduangrat N., Chumnanpuen P., Shoombuatong W. PSR-MAPMS: A new approach for the interpretable prediction of myelin autoantigenic peptides in multiple sclerosis using multi-source propensity scores. Protein Science Vol.34 No.8 (2025). doi:10.1002/pro.70010 Retrieved from: https://repository.li.mahidol.ac.th/handle/123456789/111370
Title
PSR-MAPMS: A new approach for the interpretable prediction of myelin autoantigenic peptides in multiple sclerosis using multi-source propensity scores
Author's Affiliation
Corresponding Author(s)
Other Contributor(s)
Abstract
Within the central nervous system, the myelin sheath is composed of elements known as myelin autoantigens that are mistakenly targeted by the immune system in multiple sclerosis (MS). This autoimmune attack leads to the destruction of myelin, resulting in the neurological symptoms characteristic of MS. Identifying myelin autoantigenic peptides is crucial for understanding the pathogenesis of MS and developing targeted therapies. Traditional approaches often struggle with the complexity and heterogeneity of biological data, making it challenging to achieve accurate predictions in a cost-effective manner. Alternatively, computational approaches that utilize sequence information can aid in the biological elucidation of peptides. In this study, we present a novel propensity score-based approach, termed PSR-MAPMS, to predict and characterize T cell-specific myelin autoantigenic peptides in MS (MAPMSs). To the extent of our knowledge, PSR-MAPMS is the first machine learning (ML)-based approach that can predict and analyze MAPMSs based solely on sequence information. In PSR-MAPMS, we generated multiple aspects of propensity scores for MAPMSs. Important propensity scores were then chosen and applied to create the final hybrid model using an ensemble learning strategy. Extensive experiments results showed that PSR-MAPMS surpasses several conventional ML-based classifiers for MAPMS prediction in both cross-validation and independent tests. In the independent test results, the accuracy, MCC, and F1 scores of PSR-MAPMS were within the ranges of 0.899–0.949, 0.800–0.899, and 0.903–949, respectively. Moreover, our estimated propensity scores can identify crucial biochemical and physicochemical properties of MAPMSs, providing valuable revelations of the fundamental biological mechanisms, which facilitates the development of more effective and personalized treatments for MS. In addition, we created a simple-to-navigate web server for PSR-MAPMS, which is publicly accessible at https://pmlabqsar.pythonanywhere.com/PSR-MAPMS.