Department of Otorhinolaryngology, Union Hospital, Tongji Medical College, Huazhong University of Science and Technology, Wuhan, 430022, China.
Sci Rep. 2023 Apr 25;13(1):6736. doi: 10.1038/s41598-023-32620-6.
Head and neck squamous cell carcinoma (HNSCC) is a heterogeneous tumor that is highly aggressive and ranks fifth among the most common cancers worldwide. Although, the researches that attempted to construct a diagnostic model were deficient in HNSCC. Currently, the gold standard for diagnosing head and neck tumors is pathology, but this requires a traumatic biopsy. There is still a lack of a noninvasive test for such a high-incidence tumor. In order to screen genetic markers and construct diagnostic model, the methods of random forest (RF) and artificial neural network (ANN) were utilized. The data of HNSCC gene expression was accessed from Gene Expression Omnibus (GEO) database; we selected three datasets totally, and we combined 2 datasets (GSE6631 and GSE55547) for screening differentially expressed genes (DEGs) and chose another dataset (GSE13399) for validation. Firstly, the 6 DEGs (CRISP3, SPINK5, KRT4, MMP1, MAL, SPP1) were screened by RF. Subsequently, ANN was applied to calculate the weights of 6 genes. Besides, we created a diagnostic model and nominated it as neuralHNSCC, and the performance of neuralHNSCC by area under curve (AUC) was verified using another dataset. Our model achieved an AUC of 0.998 in the training cohort, and 0.734 in the validation cohort. Furthermore, we used the Cell-type Identification using Estimating Relative Subsets of RNA Transcripts (CIBERSORT) algorithm to investigate the difference in immune cell infiltration between HNSCC and normal tissues initially. The selected 6 DEGs and the constructed novel diagnostic model of HNSCC would make contributions to the diagnosis.
头颈部鳞状细胞癌（HNSCC）是一种异质性肿瘤，具有高度侵袭性，在全球最常见癌症中排名第五。尽管尝试构建诊断模型的研究在 HNSCC 中存在缺陷。目前，诊断头颈部肿瘤的金标准是病理学，但这需要进行创伤性活检。对于这种高发肿瘤，仍然缺乏非侵入性测试。为了筛选遗传标志物并构建诊断模型，采用随机森林（RF）和人工神经网络（ANN）的方法。从基因表达综合数据库（GEO）数据库中获取 HNSCC 基因表达数据；我们总共选择了三个数据集，我们将两个数据集（GSE6631 和 GSE55547）结合起来筛选差异表达基因（DEGs），并选择另一个数据集（GSE13399）进行验证。首先，通过 RF 筛选出 6 个 DEGs（CRISP3、SPINK5、KRT4、MMP1、MAL、SPP1）。随后，ANN 用于计算 6 个基因的权重。此外，我们创建了一个诊断模型，并将其命名为 neuralHNSCC，并使用另一个数据集验证 neuralHNSCC 的曲线下面积（AUC）性能。我们的模型在训练队列中获得了 0.998 的 AUC，在验证队列中获得了 0.734 的 AUC。此外，我们使用基于相对 RNA 转录本估计的细胞类型识别（CIBERSORT）算法初步研究 HNSCC 和正常组织之间免疫细胞浸润的差异。选定的 6 个 DEGs 和构建的新型 HNSCC 诊断模型将有助于诊断。
