Comment on “Privacy-preserving multicenter validation of the RMODscore” Authors: Sudhakar Sengan, Nurmuhammad Mamazulunov, Iroda Khujamkulova Published: 2026 Dear Editor, Wu et al. (2026) developed a A novel RNA modification-based risk score as a predictive biomarker for melanoma prognosis and immunotherapy response. They analyzed 84 regulators of RNA modification, performed LASSO-based feature selection and multivariate Cox regression, and validated the 10-gene model in four independent immunotherapy datasets. These results are a promising basis for biomarker development. Prospective validation using clinical and genomic data from multiple institutions is required before the RMODscore can be considered for clinical translation. Limitations of retrospective validation External validation using Gene Expression Omnibus (GEO) cohorts demonstrates transportability to historical cohorts, but does not capture operational variability when hospitals store genomic and clinical records in-house. Such variability may be due to differences in sequencing technology, preprocessing workflows, therapeutic regimens, demographic composition, censoring methods and follow-up schedules, all of which may affect calibration and discrimination. Transcriptomic profiles can be sensitive and potentially re-identifiable, which may limit their centralized sharing. These factors may prevent evaluation of otherwise promising signatures at sufficient scale. Privacy-preserving federated framework A federated validation framework would allow participating melanoma centers to collaboratively assess a pre-determined survival model, with patient-level data staying at the local centers. Each center could apply a common data-harmonization protocol, compute the fixed RMODscore locally, and provide privacy-protected summary statistics for discrimination, calibration, clinical utility, and subgroup performance assessment. In case of recalibration or coefficient updating, secure aggregation could be used to aggregate protected parameter updates. In addition, one should also consider the use of differentially private gradient clipping and calibrated noise addition, because federated learning alone does not eliminate the possibility of information leakage from model updates. The privacy parameters epsilon (ε) and delta (δ) should be reported explicitly, as should their impact on predictive performance and clinical utility (Sheller et al. 2020). The evaluation is proposed to be: local only, standard federated, federated with differential privacy, and (if allowed by institutional governance) centrally pooled models. Performance measures, such as Harrell’s C-index, time-dependent area under the curve, integrated Brier score, calibration slope, and decision-curve net benefit, should be prespecified. Leave-one-center-out validation would provide a stronger assessment of geographic transportability than holding out only one selected center. Subgroup analyses by age, sex, disease stage, ancestry, and immunotherapy regimen should evaluate potential performance heterogeneity. Privacy robustness should also be assessed using established membership-inference and gradient-reconstruction tests. Report confidence intervals and center-specific performance distributions because aggregate estimates may conceal poor performance in smaller participating centers. Pathway to multicenter clinical translation Such an investigation would extend the evidence supporting the RMODscore beyond retrospective validation using public datasets. In a multi-institutional medical-imaging study, federated learning achieved model quality close to that obtained through centralized data sharing by Sheller et al. (2020). Multicenter oncology studies have also demonstrated the feasibility of federated approaches for treatment-response prediction by Liu et al. (2024), Ogier du Terrail et al. (2023), while differential privacy has been investigated in federated multi-omics survival analysis by Wen and Li (2025). A privacy-preserving federated assessment could therefore determine whether the RMODscore remains reliable across genuine institutional differences while providing a governance-conscious pathway toward prospective clinical validation. In conclusion, the RMODscore is a promising research biomarker that warrants further prospective evaluation. Multicenter federated validation could strengthen its clinical evidence by jointly assessing discrimination, calibration, clinical utility, subgroup performance, privacy risks, and the privacy–utility trade-off introduced by differential privacy. Such validation should precede any claim of clinical readiness or routine implementation. Acknowledgements Not applicable. Author contributions S.S. contributed to conceptualization, literature review, critical analysis, interpretation of the published evidence, and writing of the original draft: N. M and I. K contributed to scientific review, methodological evaluation, critical revision, manuscript editing, supervision, and final approval of the manuscript. Both authors reviewed and approved the final manuscript. Data availability No datasets were generated or analysed during the current study. Declarations Conflict of interest The authors declare no competing interests. Ethical approval Not applicable. This manuscript is a commentary on a previously published study and did not involve human participants or animals. Informed consent Not applicable. Generative AI disclosure Generative AI tools were used solely for language refinement and formatting assistance. The authors independently performed all scientific interpretation, critical appraisal, and manuscript preparation. Footnotes The online version of the original article can be found at https://doi.org/10.1007/s00432-026-06557-y. Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. References Associated Data Data Availability Statement No datasets were generated or analysed during the current study.