Page 97 - EJMO-9-3

Eurasian Journal of
Medicine and Oncology Novel senescence-based melanoma risk model

stability with additional clusters. This approach ensured that RNA-sequencing data. CIBERSORT utilizes a pre-
28
the identified subtypes were both biologically meaningful defined gene expression signature matrix derived from
and statistically robust. To assess the clinical relevance of isolated immune cell populations to infer cell-type-specific
these subtypes, Kaplan-Meier (KM) survival curves were abundances in complex tissue samples. This approach
generated, and the log-rank test was performed using is particularly advantageous for dissecting the intricate
the “survival” R package (https://CRAN.R-project.org/ immune composition of tumors, as it provides a high-
package=survival). The stability of the clustering results resolution view of immune infiltration patterns that are
was further confirmed by examining the distribution of critical for understanding tumor-immune interactions
key clinical variables, such as tumor stage and patient age, and predicting therapeutic responses. To assess variations
across the two subtypes. in immune cell infiltration across predefined patient
subgroups (e.g., high-risk vs. low-risk or cluster 1 vs.
2.4. Pathway enrichment analysis and assessment of cluster 2), the Wilcoxon rank-sum test was applied, a
activation score non-parametric statistical method suitable for comparing
Differentially expressed genes between the two subtypes two independent groups with non-normally distributed
were identified using the “limma” R package. To ensure data. A significance threshold of p<0.05 was applied to
statistical rigor, p-values were adjusted using the Benjamini- identify statistically significant differences in immune cell
Hochberg (BH) correction to control the false discovery abundances. This analysis revealed pronounced disparities
+
rate, with a significance threshold of p<0.05. In addition, in key immune cell populations, including CD8 T cells, NK
the absolute log2 fold change (|log2FC|) was used with a cells, and M2 macrophages, underscoring the heterogeneity
cutoff of |log2FC| > 0.5, indicating substantial differences of immune responses in melanoma. To investigate the
in gene expression between the two subtypes. Pathway relationship between the expression levels of prognostic
information was retrieved from the Kyoto Encyclopedia senescence-related genes and the enrichment of immune
of Genes and Genomes database. Enrichment analysis cells, Pearson correlation coefficients were computed
was then performed for the significantly upregulated for each gene-immune cell pair. Genes with a Pearson
genes in cluster 2, which exhibited characteristics of an correlation coefficient >0.4 and p<0.05 were considered
immunoactivated subtype. This analysis was performed significantly correlated with immune cell infiltration.
using the “clusterProfiler” R package, providing a
systematic approach to identify key biological pathways 2.6. Risk model construction and validation
associated with immune activation. Functional annotation To enhance the precision of identifying key prognostic
and enrichment analysis facilitated a more thorough senescence-related genes, Lasso-Cox regression was
insight into the molecular mechanisms underlying the employed – a regularization technique that combines
immunological differences between melanoma subtypes. feature selection and shrinkage to improve model
To ensure robustness and reduce false positives, a threshold interpretability and predictive accuracy. This analysis
of adjusted p-value (BH) <0.05 was applied to identify was carried out on the training set using the “glmnet” R
significantly enriched pathways. Hallmark pathways were package, with the optimal penalty parameter (lambda.
obtained from MSigDB (https://www.gsea-msigdb.org/ min) determined through 10-fold cross-validation. The
gsea/msigdb). Gene set variation analysis (GSVA, https:// lambda.min value represents the penalty parameter that
bioconductor.org/packages/GSVA/), a non-parametric, minimizes the cross-validated prediction error, ensuring
unsupervised approach, which transforms gene expression that the selected genes are both statistically robust and
data into gene set activity scores for each sample, was biologically relevant. By reducing the coefficients of less
applied to assess overall pathway activity. Pathways with significant genes to zero, Lasso regression effectively
significantly distinct activation scores between the two reduces overfitting and identifies a concise set of genes
subtypes were selected using the “limma” R package, with with the strongest prognostic significance. Following
a cutoff of adjusted p-value (BH) <0.05 and |log2FC| >0.5. Lasso regression, multivariate Cox regression analysis
was performed using the “survival” R package to develop
2.5. Evaluation and comparison of immune cell the final risk model. Each selected gene was assigned a
infiltration across samples regression coefficient, reflecting its individual contribution
To comprehensively characterize the immune landscape to patient survival. The risk score for each sample was
within the TME of SKCM and external GEO datasets, the determined using the following formula:
CIBERSORT algorithm was employed – a deconvolution-
based computational tool designed to assess the relative RiskScore = n ( ∑ coef ×exp )
proportions of 22 distinct immune cell types from bulk i =1 i i

Volume 9 Issue 3 (2025) 89 doi: 10.36922/ejmo.8574