Identification of gene expression signatures for psoriasis classification using machine learning techniques

Abstract

Psoriasis classification requires the accurate identification of the lesional types for the early and effective diagnosis and it is worth interesting that the normal and psoriasis cell tissues exhibit different gene expression. Therefore, gene expression data is an effective source for psoriasis classification and there is a challenge regarding the selection of suitable gene signatures for its purpose. In this present study, the gene expression-based microarray data were used and 35 expression features linked with psoriasis were utilized to feed into our machine learning model. Overall, the performance of our model based on 35 mentioned-above features surpassed that of other state-of-the-art classifiers with an average accuracy of 98.3%, recall of 98.6%, and precision of 98% in 5-fold cross-validation tests. We also validate our model on two different sets of psoriasis and the performance results are significant. These results have suggested that our 35 expression signatures have been identified as key features for classifying samples between lesion and non-lesion. More specifically, the expression levels of few genes i.e., FABP5, TGM1, or BCAR3 are discovered as newly potential biomarkers for psoriasis classification and treatment with high confidence. This study, therefore, could shed light on developing the prediction models for psoriasis classification and treatment using gene expression profiles.