Skip to content

Investigating the Signature Genes of Atopic Dermatitis and Psoriasis

Highlights
  • Atopic dermatitis lesional: NEMP2, SOX21-AS1, RP11-775C24.5, C5orf46;
  • Atopic dermatitis non-lesional: ASAH1, NET1, LINC01431;
  • Psoriasis lesional: ZBTB11, LINC01431, SLC6A16, STYK1;
  • Psoriasis non-lesional: RP11-487I5.4, RP11-332H18.3, GPLD1.

Lay Summary

Atopic Dermatitis (AD) and Psoriasis (PSO) are skin conditions affecting approximately 20% and 2% of the global population respectively. Although these diseases are well known to physicians and have been for some time, their causes have eluded scientists up until fairly recently. In this project, we sought to uncover some of the inherent genetic causes of these conditions by identifying key genetic markers that may make people more susceptible to the diseases. To do this, a dataset containing the genes of both healthy and unhealthy patients was scrutinised using a variety of Machine Learning techniques. The results of this study have highlighted a number of genes that may place people at a greater disposition to these conditions. Further research would be helpful in corroborating the results of this study, however, if successful, the findings could form the basis of a pre-diagnostic tool for determining the likelihood of a patient succumbing to these skin conditions in their lifetime.

Abstract

Atopic Dermatitis (AD) and Psoriasis (PSO) are inflammatory skin diseases with complex molecular pathogenesis and a range of phenotypes. In this project, signature genes for AD lesional, AD non-lesional, PSO lesional and PSO non-lesional conditions were identified through the application of Machine Learning (ML) methods. Five supervised ML algorithms were implemented with the task of distinguishing between the conditions based on a dataset with 147 patient samples, each having values for 31,262 unique gene expression signatures. Four algorithms (Extreme Gradient Boosting, Support Vector Machine, Logistic Regression and Linear Discriminant Analysis) produced Test Accuracies above 70%, with the best performing algorithm being Extreme Gradient Boosting, which yielded a Test Accuracy of 79%. The algorithms were then used to output weightings for each gene which lead to the most important genes being identified for the diseases. The AD-related genes were NEMP2, SOX21-AS1, RP11-775C24.5, C5orf46, ASAH1, NET1, LINC01431; PSO-related genes were ZBTB11, LINC01431, SLC6A16, STYK1, RP11-487I5.4, RP11-332H18.3 and GPLD1. Several of the identified signature genes had links to cancer, which potentially supports the known association between cancer and AD and PSO. Further insights into the signature genes uncovered in this project could lead to a better understanding of the diseases, novel early intervention strategies, and more targeted therapy for AD and PSO.


Keywords: Atopic Dermatitis, Psoriasis, Signature Genes, Machine Learning