Machine learning-based screening for biomarkers of psoriasis and immune cell infiltration

Background: Psoriasis is a chronic immune-mediated skin disease. However, the pathogenesis is not yet well established. Objectives: This study aimed to screen psoriasis biomarker genes and analyse their significance in immune cell infiltration. Materials & Methods: GSE13355 and GSE14905 datasets were downloaded from Gene Expression Omnibus (GEO) as training groups to establish the model. GSE30999 obtained from GEO was used to validate the model. Differential expression and multiple enrichment analyses were performed on 91 psoriasis samples and 171 control samples from the training group. The "LASSO" regression model and support vector machine model were used to screen and verify genes implicated in psoriasis. Genes with an area under the ROC curve >0.9 were selected as candidate biomarkers and verified in the validation group. Differential analysis of immune cell infiltration was performed on psoriasis and control samples using the "CIBERSORT" algorithm. Correlation analyses between the screened psoriasis biomarkers and 22 types of immune cell infiltration were performed. Results: In total, 101 differentially expressed genes were identified, which were mainly shown to be involved in regulating cell proliferation and immune functions. Three psoriasis biomarkers, BTC, IGFL1, and SERPINB3, were identified using two machine learning algorithms. These genes showed high diagnostic value in training and validation groups. The proportion of immune cells during immune infiltration differed between psoriasis and control samples, which was associated with the three biomarkers. Conclusion: BTC, IGFL1, and SERPINB3 are associated with the infiltration of multiple immune cells, and may therefore be used as biomarkers for psoriasis.