Abstract: Objective To screen the pathogenic hub genes of psoriasis by machine learning algorithm and to explore the potential mechanism of immune cell subtype recruitment. 
Methods Transcriptome sequencing data sets of psoriasis lesional samples (LS) (n=44) and non-lesional tissue samples (NLS) (n=44) were downloaded from public database (GSE142582,GSE161683 GSE121212). We screened differentially expressed genes (DEGs) between LS and NLS. The machine learning algorithm was applied to screen the hub genes, and ssGSEA was used to calculate the immune cell infiltration (ICI) level in the samples. The correlation between hub genes expression level and ICI level was analyzed. 
Results A total of 151 DEGs were screened out in this study. Four hub genes were identified by WGCNA and LASSO regression analysis, which were CCL20, TNFRSF17, ALDH3A1 and EPN3. The four genes had good diagnostic efficacy. CCL20 and TNFRSF17 positively regulated the ICI in LS, while ALDH3A1 and EPN3 negatively regulated ICI in NLS. 
Conclusion Four genes, including CCL20, ALDH3A1, EPN3 and TNFRSF17, have good diagnostic efficacy in psoriasis lesions. CCL20 and TNFRSF17 are positively correlated with ICI level in LS, which may recruit immune cells. However, ALDH3A1 and EPN3 may negatively regulate immune response and play a protective role in skin. CCL20, ALDH3A1, EPN3 and TNFRSF17 may be the potential biomarkers and therapeutic targets of psoriasis. 

Key words: psoriasis, Th17 cells, hub gene