Identifying biomarkers for methotrexate efficacy in rheumatoid arthritis: a machine learning approach to whole-blood transcriptomic analysis
Abstract
Objectives: Methotrexate (MTX) is the first-line treatment for rheumatoid arthritis (RA), yet inadequate response is reported in 30–40% of patients. Predicting MTX response early could enable more personalized and effective treatment. This study aimed to identify biomarkers predictive of MTX response at 6 months through whole-blood transcriptomic signatures using machine learning.
Methods: RNA sequencing data were generated in whole-blood samples taken from 100 MTX-naive RA patients at baseline (pre-treatment) and following 4 weeks post treatment with MTX. Machine learning models were trained to classify MTX response following 6 months of treatment using gradient boosted trees and interpreted using SHapley Additive exPlanations (SHAP) values to identify predictive genes.
Results: Machine learning models trained on baseline and 4-weeks data achieved AUCs of 0.89 and 0.90, respectively. Stability of SHAP values showed that the baseline model was generally more stable, and therefore potentially more generalizable. Key predictive genes at baseline, which were downregulated in responders, included CAV1, LCN12 and GLB1L.
Conclusion: Blood-based gene expression profiling at baseline and after 4 weeks of MTX treatment can predict treatment response with high confidence revealing relevant gene pathways and candidate gene targets but requires independent validation. These findings highlight the potential for transcriptomic biomarkers to inform early treatment decision in RA, supporting precision medicine approaches.
Related
- An Active Cytomegalovirus Infection and Epstein-Barr Virus Reactivation Complicated by Cytomegalovirus Meningitis and Virus-Associated Hemophagocytic Lymphohistiocytosis during Methotrexate and Upadacitinib Therapy in a Patient with Rheumatoid Arthritis
- Contemporary Guideline Perspectives on Methotrexate and Biologic/Targeted Synthetic Disease-Modifying Antirheumatic Drugs (DMARDs) in Rheumatoid Arthritis-Associated Interstitial Lung Disease
- Machine Learning Models to Predict Methotrexate Treatment Response in Rheumatoid Arthritis: A Scoping Review
- Patient perspectives on simultaneous methotrexate-folic acid intake to improve adherence in rheumatoid arthritis: a cross-sectional survey.
- Isolated Medial Ankle Tenosynovitis of the Flexor Digitorum and Posterior Tibialis Tendons: A Case Report of the Role of Early Musculoskeletal Ultrasound in Identifying Anti-citrullinated Protein Antibody (ACPA)-Positive At-Risk Rheumatoid Arthritis
BibTeX
@article{yap2026identifying,
title = {{Identifying biomarkers for methotrexate efficacy in rheumatoid arthritis: a machine learning approach to whole-blood transcriptomic analysis}},
author = {Chuan Fu Yap and Nisha Nair and Suzanne M M Verstappen and Kimme Hyrich and Guillermo Barturen and María Rivas-Torrubia and Marta Gut and Ivo Gut and Marta E Alarcón Riquelme and Anne Barton and Darren Plant and Ade Adebajo and Ade Adebajo and Khalid Ahmed and Atheer Al-Ansari and Roshan Amarasena and Marwan Bukhari and Margaret Callan and Easwaradhas G Chelliah and Hector Chinoy and Annie Cooper and Bhaskar Dasgupta and Martin Davis and James Galloway and Andrew Gough and Michael Green and Nicola Gullick and Jennifer Hamilton and Waji Hassan and Samantha Hider and Kimme Hyrich and Sanjeet Kamath and Susan Knight and Suzanne Lane and Martin Lee and Sarah Levy and Lizzy Macphie and Christopher Marguerie and Tarnya Marshall and Catherine Mathews and Frank McKenna and Sophia Naz and Mark Perry and Louise Pollard and Brian Quilty and Lindsay Robertson and Dipak Roy and Paul Sanders and Vadivelu Saravanan and David Scott and Gillian Smith and Richard Smith and Deborah Symmons and Lee-Suan Teh and Nick Viner},
journal = {Rheumatology (Oxford, England)},
year = {2026},
month = oct,
doi = {10.1093/rheumatology/keag560},
url = {https://doi.org/10.1093/rheumatology/keag560},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - Identifying biomarkers for methotrexate efficacy in rheumatoid arthritis: a machine learning approach to whole-blood transcriptomic analysis
AU - Chuan Fu Yap
AU - Nisha Nair
AU - Suzanne M M Verstappen
AU - Kimme Hyrich
AU - Guillermo Barturen
AU - María Rivas-Torrubia
AU - Marta Gut
AU - Ivo Gut
AU - Marta E Alarcón Riquelme
AU - Anne Barton
AU - Darren Plant
AU - Ade Adebajo
AU - Ade Adebajo
AU - Khalid Ahmed
AU - Atheer Al-Ansari
AU - Roshan Amarasena
AU - Marwan Bukhari
AU - Margaret Callan
AU - Easwaradhas G Chelliah
AU - Hector Chinoy
AU - Annie Cooper
AU - Bhaskar Dasgupta
AU - Martin Davis
AU - James Galloway
AU - Andrew Gough
AU - Michael Green
AU - Nicola Gullick
AU - Jennifer Hamilton
AU - Waji Hassan
AU - Samantha Hider
AU - Kimme Hyrich
AU - Sanjeet Kamath
AU - Susan Knight
AU - Suzanne Lane
AU - Martin Lee
AU - Sarah Levy
AU - Lizzy Macphie
AU - Christopher Marguerie
AU - Tarnya Marshall
AU - Catherine Mathews
AU - Frank McKenna
AU - Sophia Naz
AU - Mark Perry
AU - Louise Pollard
AU - Brian Quilty
AU - Lindsay Robertson
AU - Dipak Roy
AU - Paul Sanders
AU - Vadivelu Saravanan
AU - David Scott
AU - Gillian Smith
AU - Richard Smith
AU - Deborah Symmons
AU - Lee-Suan Teh
AU - Nick Viner
PY - 2026
DA - 2026/10/08
JO - Rheumatology (Oxford, England)
DO - 10.1093/rheumatology/keag560
UR - https://doi.org/10.1093/rheumatology/keag560
AB - Objectives: Methotrexate (MTX) is the first-line treatment for rheumatoid arthritis (RA), yet inadequate response is reported in 30–40% of patients. Predicting MTX response early could enable more personalized and effective treatment. This study aimed to identify biomarkers predictive of MTX response at 6 months through whole-blood transcriptomic signatures using machine learning. Methods: RNA sequencing data were generated in whole-blood samples taken from 100 MTX-naive RA patients at baseline (pre-treatment) and following 4 weeks post treatment with MTX. Machine learning models were trained to classify MTX response following 6 months of treatment using gradient boosted trees and interpreted using SHapley Additive exPlanations (SHAP) values to identify predictive genes. Results: Machine learning models trained on baseline and 4-weeks data achieved AUCs of 0.89 and 0.90, respectively. Stability of SHAP values showed that the baseline model was generally more stable, and therefore potentially more generalizable. Key predictive genes at baseline, which were downregulated in responders, included CAV1, LCN12 and GLB1L. Conclusion: Blood-based gene expression profiling at baseline and after 4 weeks of MTX treatment can predict treatment response with high confidence revealing relevant gene pathways and candidate gene targets but requires independent validation. These findings highlight the potential for transcriptomic biomarkers to inform early treatment decision in RA, supporting precision medicine approaches.
SN - 1462-0324
SN - 1462-0332
ER -