Development and validation of a predictive gene signature

Using the TCGA-KIRC dataset as a training cohort, for the identified hub genes, we performed a univariate Cox regression survival analysis to identify prognostic CRGs, applying a significance threshold of P < 0.05. To mitigate overfitting and optimize the model performance, we conducted Least Absolute Shrinkage and Selection Operator (LASSO) Cox regression analysis with 10-fold cross-validation and 1000 iterations. We used the “glmnet” package (ver. 4.1-6) to perform this analysis on the identified genes. Subsequently, the CAF score was calculated by the following method (Equation 1):

According to the median CAF score, we classified the 533 ccRCC samples into two categories: low-CAF and high-CAF groups. Scatter plots elucidated the correlation between CAF scores and survival status. Both PCA and t-Distributed Stochastic Neighbor Embedding (t-SNE) were employed to evaluate the capability of the risk model to distinguish between the low-CAF and high-CAF groups. To assess the differential overall survival between the two groups, we conducted a survival analysis utilizing Kaplan-Meier curves and log-rank tests. We also plotted time-dependent Receiver Operating Characteristic (ROC) curves to gauge the model’s prognostic accuracy. For further validation, we also applied the CAF risk model to the E-MTAB-1980 dataset.