Establishment and validation of evaluation models for post-inflammatory pigmentation abnormalities

Abstract

Post-inflammatory skin hyper- or hypo-pigmentation is a common occurrence with unclear etiology. There is currently no reliable method to predict skin pigmentation outcomes after inflammation. In this study, we analyzed the 5 GEO datasets to screen for inflammatory-related genes involved in melanogenesis, and used candidate cytokines to establish different machine learning (LASSO regression, logistic regression and Random Forest) models to predict the pigmentation outcomes of post-inflammatory skin. Further, to further validate those models, we evaluated the role of these candidate cytokines in pigment cells. We found that IL-37, CXCL13, CXCL1, CXCL2 and IL-19 showed high predictive value in predictive models. All models accurately classified skin samples with different melanogenesis-related gene scores in the training and testing sets (AUC>0.7). Meanwhile, we mainly evaluated the effects of IL-37 in pigment cells, and found that it increased the melanin content and expression of melanogenesis-related genes (MITF, TYR, TYRP1 and DCT), also enhanced tyrosinase activity. In addition, CXCL13, CXCL1, CXCL2 and IL-19 could down-regulate the expression of several melanogenesis-related genes. In conclusion, evaluation models basing on machine learning may be valuable in predicting outcomes of post-inflammatory pigmentation abnormalities. IL-37, CXCL1, CXCL2, CXCL13 and IL-19 are involved in regulating post-inflammatory pigmentation abnormalities.

Keywords: IL-37; evaluation models; machine learning; melanogenesis; post-inflammatory pigmentation abnormalities.