Aim. In this paper, we aimed to develop and validate a risk prediction method using independent prognosis genes selected robustly in prostate cancer. Method. We considered 723 samples obtained from TCGA (the Cancer Genome Atlas), GSE46602, and GSE21032. Prostate cancer prognosis-related genes with were selected using Univariable Cox regression analysis. We then built the lowest AIC (Akaike information criterion score) optimal gene model using the “Rbsurv” package in TCGA train set. The coefficients were obtained by Multivariable Cox regression analysis. We named the new prognosis method CMU5. The CMU5 risk score was verified in TCGA test set, GSE46602, and GSE21032. Results. FAM72D, ARHGAP33, TACR2, PLEK2, and FA2H were identified as independent prognosis factors in prostate cancer patients. We built the computing model as follows: CMU5 risk score = 1.158FAM72D + 1.737ARHGAP33 − 0.737TACR2 − 0.651PLEK2 − 0.793FA2H. The AUC of DFS was 0.809 in the train set (274 samples), 0.710 in the test set (273 samples), and 0.768 in the complete set (547 samples). The benign prediction capacity of CMU5 was verified by GSE46602 (36 samples; ) and GSE21032 GPL5188 (140 samples; ). Using the cut-off point of 2.056, a significant difference was shown between high- and low-risk groups. Conclusion. A prognosis-related risk score formula named CMU5 was built and verified, providing reliable prediction of prostate cancer outcome. This signature might provide a basis for individualized treatment of prostate cancer.

中文翻译：

---

确定前列腺癌中稳健的五基因风险模型：基于稳健的可能性生存分析。

瞄准。在本文中，我们旨在开发和验证使用在前列腺癌中稳健选择的独立预后基因的风险预测方法。方法。我们考虑了从 TCGA（癌症基因组图谱）、GSE46602 和 GSE21032 获得的 723 个样本。前列腺癌预后相关基因使用单变量 Cox 回归分析选择。然后，我们使用 TCGA 训练集中的“Rbsurv”包构建了最低 AIC（Akaike 信息标准分数）最优基因模型。通过多变量Cox回归分析获得系数。我们将新的预后方法命名为 CMU5。CMU5 风险评分在 TCGA 测试集、GSE46602 和 GSE21032 中得到验证。结果。FAM72D、ARHGAP33、TACR2、PLEK2和FA2H被确定为前列腺癌患者的独立预后因素。我们构建的计算模型如下：CMU5 风险评分 = 1.158 FAM72D  + 1.737 ARHGAP33  - 0.737 TACR2  - 0.651PLEK2  - 0.793 FA2H。DFS 的 AUC 在训练集（274 个样本）中为 0.809，在测试集（273 个样本）中为 0.710，在完整集（547 个样本）中为 0.768。CMU5的良性预测能力通过GSE46602（36个样本；)和 GSE21032 GPL5188（140 个样本；）。使用 2.056 的截止点，高风险组和低风险组之间显示出显着差异。结论。建立并验证了一个名为 CMU5 的预后相关风险评分公式，为前列腺癌预后提供了可靠的预测。该特征可能为前列腺癌的个体化治疗提供基础。