稀疏自动编码器在甲状腺乳头状癌分型中的应用*

秦阳 ,  任月 ,  兰宁 ,  张永超 ,  刘毅蓉 ,  赵晋芳 ,  罗天娥

中国卫生统计 ›› 2026, Vol. 43 ›› Issue (5) : 693 -698.

PDF (2291KB)
中国卫生统计 ›› 2026, Vol. 43 ›› Issue (5) : 693 -698. DOI: 10.11783/j.issn.1002-3674.2026.05.009
论著

稀疏自动编码器在甲状腺乳头状癌分型中的应用*

作者信息 +

Application of Sparse Autoencoder in the Subtype Classification of Papillary Thyroid Carcinoma

Author information +
文章历史 +
PDF (2345K)

摘要

目的 探索稀疏自动编码器(sparse autoencoder,SAE)在甲状腺乳头状癌(papillary thyroid carcinoma, PTC)分子分型中的应用价值。方法 整合PTC 中分型不明确型(not otherwise specified,NOS)患者的mRNA、miRNA、 DNA 甲基化数据。采用自动编码器(autoencoder,AE)和SAE 进行特征提取,用k-means 聚类并获取对应标签,用轮廓系 数(silhouette coefficient,SC)、方差比准则(calinski-harabasz index,CH)和 Davies-Bouldin 指数(Davies-Bouldin index, DBI)来评估模型的聚类效果。构建随机生存森林模型获取影响预后的重要基因。结果 在SAE-k-means 的结果中,最 佳聚类数为2,将PTC 中的NOS 患者分成两个亚型,即subtype1 和subtype2。两个亚型之间的生存差异具有统计学意义 (χ2=5.36,P=0.021),subtype2 预后较差。在 subtype1 中,CXCL5 和DGKI 等基因有重要的临床意义;在 subtype2 中, ATP2C2、EHF、PON2 等基因有重要的临床意义。结论 SAE 在聚类和预后方面的性能都优于AE,有效识别了不同的亚 型,寻找到了相对重要的影响预后的基因,为PTC分子分型提供了新思路,为治疗和预后评估提供了理论支持。

Abstract

Objective To explore the application value of sparse autoencoder (SAE) in molecular subtyping of papillary thyroid carcinoma (PTC).Methods To integrate the data of mRNA, miRNA, and DNA methylation of patients with not otherwise specified (NOS) in PTC. Autoencoder (AE) and SAE are used to extract features, and k-means is used to cluster and obtain corresponding labels. The clustering effect of the model is evaluated by the silhouette coefficient (SC), calinski-harabasz index (CH), and Davies-Bouldin index (DBI). A random survival forest model is constructed to obtain the important genes affecting the prognosis.Results In the results of SAE-k-means, the optimal number of clusters is 2, which divides NOS patients in PTC into two subtypes, subtype1, and subtype2. The survival difference between the two subtypes is statistically significant (χ2=5. 36, P=0. 021), with subtype2 having a poor prognosis. In subtype1, CXCL5 and DGKI have important clinical significance. In subtype2, genes such as ATP2C2, EHF, and PON2 have important clinical significance. Conclusion SAE is superior to AE in clustering and prognosis, effectively identifying different subtypes and finding relatively important genes that affect prognosis, which provides a new idea for PTC molecular typing and theoretical support for treatment and prognosis evaluation.

关键词

稀疏自动编码器 / 多组学整合 / 分子分型 / 随机生存森林

Key words

Sparse autoencoder / Multi-omics integration / Molecular subtyping / Random survival forest

引用本文

引用格式 ▾
秦阳,任月,兰宁,张永超,刘毅蓉,赵晋芳,罗天娥. 稀疏自动编码器在甲状腺乳头状癌分型中的应用*[J]. 中国卫生统计, 2026, 43(5): 693-698 DOI:10.11783/j.issn.1002-3674.2026.05.009

登录浏览全文

4963

注册一个新账户 忘记密码

参考文献

[1]

魏佳. 基于循环肿瘤 DNA 的甲状腺癌分子诊断技术研究[D].长春:吉林大学, 2022.

[2]

赵海月, 宋旸, 沈迪寒, 等. 甲状腺癌的分型及其分子发病机制研究进展[J].中国体视学与图像分析, 2017, 22(4): 465-471.

[3]

徐昭阳, 刘改琴, 王彤, 等. 基于CiteFuse 的肺腺癌单细胞多组学数据整合细胞类型识别[J].中国卫生统计, 2026, 43(3):351-358.

[4]

高振博. 基于多组学数据的聚类方法研究[D].大连:大连理工大学, 2020.

[5]

Chaudhary K, Poirion OB, Lu L, et al. Deep learning—based multi—omics integration robustly predicts survival in liver Cancer[J].Clinical Cancer Research, 2018, 24(6): 1248-1259.

[6]

李淼, 罗天娥, 郭强, 等. 随机生存森林模型在肺癌患者预后分析中的应用[J].中国卫生统计, 2021, 38(3): 327-331.

[7]

Titov SE, Ivanov MK, Karpinskaya E V, et al. miRNA profiling, detection of BRAF V600E mutation and RET—PTC1 translocation in patients from Novosibirsk oblast (Russia) with different types of thy—roid tumors[J].BMC Cancer, 2016, 16(1): 201.

[8]

Cao ZX, Weng X, Huang JS, et al. Receptor— ligand pair typing and prognostic risk model for papillary thyroid carcinoma based on single—cell sequencing[J].Frontiers in Immunology, 2022, 13: 902550.

[9]

Feldner—Busztin D, Firbas Nisantzis P, Edmunds SJ, et al. Dealing with dimensionality: the application of machine learning to multi—omics data[J].Bioinformatics, 2023, 39(2): btad021.

[10]

Madhumita, Paul S. Capturing the latent space of an Autoencoder for multi—omics integration and cancer subtyping[J].Computers in Biology and Medicine, 2022, 148: 105832.

[11]

Cui D, Zhao Y, Xu J. Activation of CXCL5—CXCR2 axis promotes proliferation and accelerates G1 to S phase transition of papillary thyroid carcinoma cells and activates JNK and p38 pathways[J].Cancer Biol Ther, 2019, 20(5): 608-616.

[12]

Cui D, Zhao Y, Xu J. Activated CXCL5—CXCR2 axis promotes the migration, invasion and EMT of papillary thyroid carcinoma cells via modulation of β —catenin pathway[J].Biochimie, 2018, 148: 1-11.

[13]

Bai M, Ke S, Yu H, et al. Key molecules associated with thyroid carcinoma prognosis: A study based on transcriptome sequencing and GEO datasets[J].Front Immunol, 2022, 13: 964891.

[14]

Zhao H, Zhang S, Shao S, et al. Identification of a Prognostic 3—Gene Risk Prediction Model for Thyroid Cancer[J].Front Endocri—nol (Lausanne), 2020, 11: 510.

[15]

Lyu Y, Sui F, Ma J, et al. Increased expression of EHF contributes to thyroid tumorigenesis through transcriptionally regulating HER2 and HER3[J].Oncotarget, 2016, 7(36): 57978-57990.

[16]

Xie F, Li L, Luo Y, et al. Long non—coding RNA LINC00488 facili—tates thyroid cancer cell progression through miR—376a—3p/PON2 [J].Biosci Rep, 2021, 41(3): BSR 20201603.

基金资助

山西省自然科学基金(201801D121210)

AI Summary AI Mindmap
PDF (2291KB)

0

访问

0

被引

详细

导航
相关文章

AI思维导图

/

〈 〉