Prediction of Psoriasis from Gene Expression Profiling Using Penalized Logistic Regression Model

Background and Objective: Psoriasis is one of the most common skin disorders in humans and is believed to have genetic foundations. The aim of this study is to identify potential genetic biomarkers for psoriasis using penalized methods. Materials and Methods: The gene chip GSE۵۵۲۰۱, which included ۷۴ individuals (۳۴ patients with psoriasis and ۳۰ healthy individuals), was obtained from GEO. Three penalized approaches were used in logistic regression, including Least Absolute Shrinkage Selection Operator, Minimax Concave Penalty, and Smoothing Clipped Absolute Deviation, to identify the most important genes associated with psoriasis. To validate the results, Random Forest was used to assess the predictive power of the selected genes in a validation dataset. Results: The analysis identified ADORA۳ and C۱۶orf۷۲ as two genes that were commonly associated with psoriasis. The independent samples t-test revealed significantly higher expression of ADORA۳ and C۱۶orf۷۲ among psoriasis cases (p<۰.۰۰۱). The area under the ROC curve for predicting psoriasis was ۰.۸۸ (۹۵% CI: ۰.۸۰-۰.۹۶) for ADORA۳ and ۰.۷۵ (۹۵% CI: ۰.۷۵-۰.۹۴) for C۱۶orf۷۲. The Random Forest analysis showed that the model using these genes had a prediction probability of ۰.۶۸ (۹۵% CI: ۰.۵۳-۰.۸۳). Conclusion: Among all the methods used, MCP outperformed other penalties, selecting a smaller subset with compatible performance. Two key genes, ADORA۳ and C۱۶orf۷۲, were found to be associated with psoriasis and were identified for further study. These genes may serve as genetic biomarkers for predicting psoriasis.

Keywords:

Gene Expression Profiling , Bioinformatics , Psoriasis , Biomarker , Prognosis , Penalized Regression