Biomarker discovery in psoriasis: an integrated featureselection and machine learning framework for identifyingdiagnostic genes
نویسندگان
1 Department of Computer Engineering, Meybod University, Meybod, Iran
2 Department of Computer Engineering, Meybod University, Meybod, Iran
3 Department of Computer Engineering, Meybod University, Meybod, Iran
4 Department of Computer Engineering, Meybod University, Meybod, Iran
doi
10.22034/ijd.2025.544542.2091چکیده
Background: Psoriasis is a chronic, immune-mediated disease characterized by inflammation and abnormal skin cell growth. The precise molecular mechanisms underlying its progression remain partially understood, highlighting the need to identify reliable gene biomarkers for early diagnosis and targeted therapies.Methods: This study presents an integrated feature selection and machine learning framework to identify key genes associated with psoriasis. Gene expression profiles were collected and preprocessed from seven GEO databases. We evaluated various combinations of evolutionary algorithms for gene selection and machine learning classifiers. The Harris Hawks Optimization (HHO) algorithm combined with the XG Boost classifier was selected due to its superior performance in maximizing prediction accuracy.Results: The hybrid HHO-XG Boost combination demonstrated superior performance, achieving an exceptional F1-score of 99.31% on the validation data. This optimized model successfully identified a set of 35 key genes highly predictive of psoriasis, including TNFRSF10C, MCOLN2, KYNU, and CD47.Conclusion: The identified 35 genes have significant potential to serve as robust diagnostic and prognostic biomarkers for psoriasis. This study validates the effectiveness of the HHO-XG Boost integrated approach for biomarker discovery in complex diseases and provides strong candidates for future functional validation and targeted drug development.
