介绍：这项研究探讨了自然杀伤（NK）细胞在肺腺癌（LUAD）中的免疫特性，以及它们对患者生存和免疫治疗反应的预测作用。材料和方法：通过评估The Cancer Genome Atlas（TCGA）数据集中的NK细胞相关通路和基因，使用一致的聚类方法对LUAD样本进行分子亚型分型。从先前的研究中获得了12个程序化细胞死亡（PCD）模式。使用最小绝对收缩和选择算子（LASSO）和Cox回归构建了风险得分预测模型。使用Gene Expression Omnibus数据库（GEO）验证了模型的稳定性。结果：我们基于NK细胞相关基因将LUAD分类为三个不同的分子亚型，C1患者预后最差，C3最佳。同源重组缺陷、纯度和倍数体积、TMB、LOH、错配体积分数在C1中最高，在C3中最低。免疫得分在C3型中最高，表明C3亚型有更大的免疫浸润。C1亚型的TIDE得分较高，表明C1亚型可能不太适合免疫治疗。总体而言，C3亚型呈现最高的PCD模式得分。我们使用ANLN、FAM83A、RHOV和PARP15这四个基因构建了LUAD风险预测模型，这两个风险组之间的免疫细胞组成和细胞周期相关通路有显著差异。C1和高组的样本对化疗药物更敏感。高组和低组的PCD得分不同。最后，我们将风险得分和临床特征相结合，以提高预测模型的性能，校准曲线和决策曲线验证了该模型的强大稳健性。结论：我们确定了三种稳定的LUAD分子亚型，并基于NK细胞相关基因构建了一个预测模型，可能对预测免疫治疗反应和患者预后有更大的应用潜力。版权所有©2023 Zhang and Zhao。

Introduction: This research explored the immune characteristics of natural killer (NK) cells in lung adenocarcinoma (LUAD) and their predictive role on patient survival and immunotherapy response. Material and methods: Molecular subtyping of LUAD samples was performed by evaluating NK cell-associated pathways and genes in The Cancer Genome Atlas (TCGA) dataset using consistent clustering. 12 programmed cell death (PCD) patterns were acquired from previous study. Riskscore prognostic models were constructed using Least absolute shrinkage and selection operator (Lasso) and Cox regression. The model stability was validated in Gene Expression Omnibus database (GEO). Results: We classified LUAD into three different molecular subgroups based on NK cell-related genes, with the worst prognosis in C1 patients and the optimal in C3. Homologous Recombination Defects, purity and ploidy, TMB, LOH, Aneuploidy Score, were the most high-expressed in C1 and the least expressed in C3. ImmuneScore was the highest in C3 type, suggesting greater immune infiltration in C3 subtype. C1 subtypes had higher TIDE scores, indicating that C1 subtypes may benefit less from immunotherapy. Generally, C3 subtype presented highest PCD patterns scores. With four genes, ANLN, FAM83A, RHOV and PARP15, we constructed a LUAD risk prediction model with significant differences in immune cell composition, cell cycle related pathways between the two risk groups. Samples in C1 and high group were more sensitive to chemotherapy drug. The score of PCD were differences in high- and low-groups. Finally, we combined Riskscore and clinical features to improve the performance of the prediction model, and the calibration curve and decision curve verified that the great robustness of the model. Conclusion: We identified three stable molecular subtypes of LUAD and constructed a prognostic model based on NK cell-related genes, maybe have a greater potential for application in predicting immunotherapy response and patient prognosis.Copyright © 2023 Zhang and Zhao.