Psoriasis is a hereditary autoimmune skin condition that is characterized by abnormally enlarged skin patches (1). The red and scaly plaques that characterize this chronic skin condition are most commonly found on the elbows, knees, scalp, and lower back (2). Psoriasis has significant psychological and social consequences, as well as imposing a high financial burden on both sufferers and healthcare systems (3). It has been suggested that many countries lack epidemiological data on this immune-mediated inflammatory skin condition. Although psoriasis is more common in adults than in children, studies have shown that it can affect people of all ages, particularly those between the ages of 20-30 and 50-60 (4). Psoriasis cases have been documented in greater numbers in developed economies (5). Psoriasis affects approximately 2 to 4% of the population in Western countries (6).
Although the origins of this non-communicable illness are unknown, several recognized risk factors include infections and skin injuries, family history, weather, certain medications, stress, smoking and exposure to secondhand smoke, obesity, and alcohol use. Psoriasis can cause severe symptoms such as burning, stinging, itching, discomfort, and skin cracking in some cases. The primary goal of psoriasis treatment is to limit cell proliferation. Psoriasis is classified into three stages: mild, moderate, and severe. Topical medications, such as calcineurin inhibitors and keratolytics, topical corticosteroids, and vitamin D analogues, can be used to treat moderate cases (7). Individuals with moderate to severe psoriasis may require more aggressive treatments, such as systemic medications, UV radiation, and rotational therapy, which involves switching medications to reduce the risk of systemic treatment toxicity over time (8). Although molecular tissue research on psoriasis is still ongoing, its genetic basis has been established. Quantitative polymerase chain reaction (qPCR) and microarray approaches have been used in comprehensive screening to identify key genes and the regulator of stem cell proliferation involved in the development of psoriasis, as well as to uncover patterns of gene expression. (4)  Psoriasis lesions are significantly associated with immune cells such as CD3+ T cells and CD11c+ dendritic cells, and the cytokines produced by these cells. In particular, tumor necrosis factor-alpha (TNFα), interferon-gamma (IFNγ), interleukin-17 (IL-17), IL-22, IL-23, IL-12, and IL-1beta are linked to the pathogenesis of psoriasis through the activation of keratinocytes and other skin cells in the dermis (8).
The selection of the most important genes from high-dimensional and correlated data is a crucial step. To assess the impact of the most important genes among a large number of genes, researchers can reduce the dimensionality of the data by using important components or employing variable selection approaches. While it has been suggested to reduce the dimensionality of data using principal component analysis and partial least squares approaches, these methods may lose some information from the data and can be difficult to interpret (9). Penalized logistic regression is a useful method for feature selection, allowing for simultaneous estimation and gene selection (10).
In this study, we employed three approaches in penalized logistic regression, including Least Absolute Shrinkage Selection Operator (LASSO), Minimax Concave Penalty (MCP), and Smoothing Clipped Absolute Deviation (SCAD), to select the most important genes associated with psoriasis.
 

The penalized logistic regression using three penalty functions identified different subsets of genes: the MCP selected three genes, the SCAD identified 15, and the LASSO approach selected 24 genes. Among the selected genes, ADORA3 and C16orf72 were common to all three penalized methods. The Bayesian Information Criterion (BIC) values for logistic regression models based on each subset of selected genes were 56, 28, and 58 for SCAD, MCP, and LASSO, respectively, indicating that the MCP selected the best subset of genes. The area under the receiver operating characteristic (ROC) curve was 1.000, 0.967, and 1.000 for SCAD, MCP, and LASSO, respectively. This suggests that the MCP could select a better subset of genes (of size 3), as its classification performance in distinguishing psoriasis patients from healthy controls was comparable to that of SCAD and LASSO with larger numbers of selected genes (15 and 24, respectively). Table 1 shows the names of the unique genes resulting from each approach, and Figure 1 displays a Venn diagram of the identified genes by the three penalized approaches.

Table 1. The resulted genes found by the three penalized approaches

LASSO: Least Absolute Shrinkage Selection Operator, MCP: Minimax Concave Penalty, SCAD: Smoothing Clipped Absolute Deviation

Figure 1. Venn diagram of identified genes by the three penalized approaches

The results of the independent samples t-test are presented in Table 2, which shows that the mean expression levels of ADORA3 and C16orf72 are significantly higher in psoriasis cases than in non-psoriasis cases (p<0.001). Furthermore, ADORA3 and C16orf72 were found to be predictors of psoriasis, with areas under the receiver operating characteristic (ROC) curve of 0.88 (95% CI: 0.80-0.96) and 0.75 (95% CI: 0.75-0.94), respectively. The ROC curve for this model is shown in Figure 2.

Table 2. Results of the independent samples t-test and the area under the ROC curve for ADORA3 and C16orf72 expressions

SD: Standard Deviation; AUC: The area under the ROC curve; CI: Confidence Interval

Figure 2. ROC curve analysis of predicting psoriasis using ADORA3 and C16orf72

Random forest was utilized to determine the predictive power of selected genes in the validation set of data. The area under the ROC curve (95% confidence interval) after applying the random forest approach was 0.68 (0.53-0.83) which showed a significant prediction by the genes in the model (p=0.034). Figure 3 demonstrates ROC curve predicting psoriasis using ADORA3 and C16orf72 in the validation dataset resulted from the random forest.

Figure 3. Random forest ROC curve predicting Psoriasis using ADORA3 and C16orf72 in the validation dataset

Discussion

Psoriasis is a chronic autoimmune/inflammatory disease with a genetic basis that affects multiple aspects of patients' health. It is associated with increased risks of various diseases, including hyperlipidemia, hypertension, coronary artery disease (CAD), diabetes (type II), stroke, and myocardial infarction. Identifying prognostic markers using state-of-the-art models is critical for early detection and treatment of psoriasis. In this study, we applied three penalizing functions in logistic regression to classify cases as psoriasis or non-psoriasis based on potential features. All the methods identified ADORA3 and C16orf72 as highly associated with the presence of psoriasis among many genes. Using a validation dataset, we applied random forest, a classification approach, and found that the selected two genes are significant predictors of psoriasis.
The protein encoded by the adenosine A3 receptor gene (ADORA3) is a G-protein-coupled receptor that is involved in the regulation of several biological functions through the inhibition of adenylate cyclase activity (12). ADORA3 has an essential role in the inhibition of inflammation (13). Our findings indicate that the expression level of ADORA3 is overexpressed in psoriatic samples, which is consistent with the results of previous studies. Stamp et al. reported a significant increase in the expression level of ADORA3 in synovial samples of patients with rheumatoid arthritis, another autoimmune disease (14). Studies have also shown that ADORA3 is present in peripheral blood mononuclear cells of patients with autoimmune inflammatory diseases, including rheumatoid arthritis (15) psoriasis, and Crohn's disease (16).  They also reported a significant reduction of IL-1b/IL-6 release in RA patients mediated by the ADORA2a and ADORA3 stimulation. Moreover, variations in the expression of ADORA subtypes have been observed and attributed to the tissues affected by the disease (17).
C16orf72, also known as Telomere Attrition and P53 Response 1 (TAPR1), plays a critical role in the negative regulation of P53 signaling induced by the loss of telomere integrity (18). This gene “exhibits a strong synthetic genetic interaction with telomerase inhibition or deletion of TERT, and appears to taper the response to p53 upon loss of telomere integrity” 19. Inflammation accelerates cellular senescence due to the shortening of telomere length, and recent studies have indicated that telomere length plays a role in the pathogenesis of various autoimmune diseases (19, 20).   
As technology has developed, genetics research has produced increasingly high-dimensional data that can be analyzed to gain insights into disease mechanisms. To avoid assessing irrelevant genes and obtain a subset of genes with high classification capabilities, dimension reduction approaches such as penalized logistic regression can be used. However, different penalized methods have their own advantages and disadvantages. LASSO is feasible for high-dimensional computations but has limitations in statistical accuracy for prediction, the number of selected genes due to the number of samples, and equality in penalizing all gene coefficients (21). SCAD provides a good goodness of fit and promotes a sparse estimate of the regression (22). In contrast, MCP utilizes similar rates of penalization to Lasso and decreases the rate to zero as the absolute value of the coefficient increases (23). In our study, these penalized logistic regression models identified the most associated genes with the presence of psoriasis.

 

Conclusion

Our findings showed that MCP outperformed other penalties by selecting a smaller subset with compatible performance. This study identified a small subset of genes, including ADORA3 and TAPR1, that can be used for the prognosis of psoriasis. Further laboratory investigations are needed to evaluate the identified genes.

 

Acknowledgements

The financial support for this study was provided by the Deputy of Research and Technology of Hamadan University of Medical Sciences (Grant NO. 9808286317).

 

Data Availability

A publicly available dataset was analyzed in this study. This data can be found at https://www.ncbi.nlm.nih.gov/geo/ (the NCBI Gene Expression Omnibus).

 

Funding

This study was funded by Hamadan University of Medical Sciences (Grant number: 9808286317).

 

Conflicts of Interest

The authors declare that there is no conflict of interests.

 

Authors' Contribution

Payam Amini: Conceptualization, methodology, software, validation, formal analysis, investigation, resources, data curation, writing-original draft, writing.
Leili Tapak: Conceptualization, methodology, software, validation, formal analysis, investigation, resources, data curation, writing-original draft, writing-review and editing, visualization, supervision, project administration, funding acquisition.
Saeid Afshar: Investigation, data curation, review and editing,
Mahlagha Afrasiabi: Investigation, data curation, software, review and editing.
MohammadKazem Ghasemi: Formal analysis, writing-review and editing.
Pedram Alirezaei: Conceptualization, methodology, validation, writing-review and editing.