Construction and validation of an immune gene pair signature to predict biochemical recurrence of prostate cancer

Abstract

Objective: Immunotherapy has become an important part of tumor therapy, and the important role of immune-related genes has also been paid more attention. This study aims to use immune-related genes to construct a model for predicting biochemical recurrence of prostate cancer. Method: The prostate cancer gene expression data downloaded from the GEO database. Then we calculated the matrix of immune gene pairs and screened out immune gene pairs related to biochemistry recurrence (BCR) by univariant COX regression. The prediction model was build by lasso regression model. The dataset of GSE54460 is used as the training set, GSE2100 as validation. The ROC curve was used to evaluate the model's predictive effectiveness. Results: In the training set, 106 BCR-related immune gene pairs were determined by univariant COX scores (p <0.001). An immune gene pair model consisting of 20 genes was constructed by lasso regression model. The optimal cutoff value was determined by the ROC curve. All the patients were divided into high-risk and low-risk groups.The survival time of the two groups without biochemical recurrence was significantly different. To further explore the role of this model, we performed enrichment analysis on genes in immune gene pairs to determine their potential signaling pathways. Conclusion: Two independent data sets was employed to construct and verify that the prediction value of immune gene model of the biochemical recurrence of prostate cancer, and the AUC reached 0.95, which has strong potential application value in clinical practice.