Establishment and validation of psoriasis evaluation models

Abstract

Psoriasis is a common inflammatory skin disease that seriously affects the patient's quality of life. The diagnosis of psoriasis is mainly based on clinical and pathological features, and the assessment depends on the psoriasis area and severity index (PASI). However, there are few reliable and accurate evaluation methods to assess lesion severity and therapeutic effects. This work identified 17 model genes from GEO datasets and established 6 psoriasis evaluation models by LASSO regression, linear regression, and random forest separately. Models were trained and evaluated in different GEO datasets. All 6 models accurately classified psoriatic lesions and non-lesional skin in training and testing data, and showed good AUC. In biologics-treated samples, the model scores were positively correlated with the severity of lesions and negatively correlated with treatment length. Thus, models have the potential to assess the therapeutic effects. In addition, the expression of model genes was examined in keratinocytes, skin of IMQ-induced psoriatic mice, and lesions of psoriasis patients. The RNA and protein levels of model genes increased in cytokine-stimulated keratinocytes and psoriatic lesions as expected. This work provides new methods to assess the lesion severity and therapeutic effects of biologics in psoriasis.