面向不平衡多组学癌症数据的特征表征算法

doi:10.13229/j.cnki.jdxbgxb.20231196

摘要/Abstract

摘要：

针对癌症疾病数据结构复杂、预测困难、数据不平衡和患者隐私保护等一系列的问题，提出了一种特征表征算法ImFeatures，解决了癌症数据的不平衡问题，丰富了样本结构。联合癌症转录组和甲基化2种组学数据作为真实样本，通过逻辑回归（LR）和随机森林（RF）2种特征选择后，得到的负样本被随机划分并结合等量的正样本，输入本文提出的特征表征模型，生成学习到关键特征信息的表征样本，以提高模型预测能力。实验结果表明，在经过特征表征后的11种常见癌症数据集上，本文提出的结合特征筛选和特征表征的算法的准确率（Acc）均超过了80.00%，其中有5种癌症的预测准确率超过了95.00%，可以有效提升癌症疾病的预测准确率。

关键词: 计算机应用技术, 特征表征, 生物信息学, 多组学数据, 特征筛选, 机器学习

Abstract:

Aiming at a series of problems such as complex data structure， difficult prediction， data imbalance， and patient privacy protection in cancer diseases， a feature representation algorithm ImFeatures was proposed to solve the problem of imbalanced cancer data and enrich the sample structure. By combining two types of cancer transcriptome and methylation data as real samples， negative samples obtained after feature selection by logistic regression and random forest were randomly divided and combined with equal numbers of positive samples. The feature representation model proposed was used to generate sample representations that learn key feature information， thereby improving the predictive ability of the model. The experimental results show that on 11 common cancer datasets after feature characterization， the accuracy （Acc） of the algorithm combining feature selection and feature representation proposed in this paper exceeds 80.00% in all cases， and five cancer types even receive accuracies over 95.00%， which can effectively improve the prediction accuracies of cancer diseases.

Key words: computer application technology, feature representation, bioinformatics, multi-omics data, feature selection, machine learning

中图分类号:

TP399

周丰丰,郭喆,范雨思. 面向不平衡多组学癌症数据的特征表征算法[J]. 吉林大学学报(工学版), 2025, 55(6): 2089-2096.

Feng-feng ZHOU,Zhe GUO,Yu-si FAN. Feature representation algorithm for imbalanced classification of multi⁃omics cancer data[J]. Journal of Jilin University(Engineering and Technology Edition), 2025, 55(6): 2089-2096.

图/表 9

图1

图2

表1

表2

表3

表4

图3

表5

表6

参考文献 21

[1]	安云鹤, 李宝明, 李越, 等. 癌症基因组测序方案制定的研究进展[J]. 中国生物工程杂志, 2014, 34(11): 9-17.
	An Yun-he, Li Bao-ming, Li Yue, et al. Progress in cancer genome-sequencing study design[J]. China Biotechnology, 2014, 34(11): 9-17.
[2]	周丰丰, 张亦弛. 基于稀疏自编码器的无监督特征工程算法BioSAE[J].吉林大学学报: 工学版, 2022, 52(7): 1645-1656.
	Zhou Feng-feng, Zhang Yi-chi. Unsupervised feature engineering algorithm BioSAE based on sparse autoencoder[J]. Journal of Jilin University (Engineering and Technology Edition), 2022, 52(7): 1645-1656.
[3]	Chen X Y, Yu Y Z W, Zheng H Y, et al. Single-cell transcriptome analysis reveals dynamic changes of the preclinical A549 cancer models, and the mechanism of dacomitinib[J]. European Journal of Pharmacology, 2023, 960: No.176046.
[4]	白天, 周春光, 王喆, 等. 代谢组学中机器学习研究进展[J]. 吉林大学学报: 信息科学版, 2008, 26(2): 163-168.
	Bai Tian, Zhou Chun-guang, Wang Zhe, et al. Advances of machine learning in metabonomics[J]. Journal of Jilin University (Information Science Edition), 2008, 26(2): 163-168.
[5]	高美虹, 尚学群. 利用人工智能预测癌症的易感性、复发性和生存期[J]. 生物化学与生物物理进展, 2022, 49(9): 1687-1702.
	Gao Mei-hong, Shang Xue-qun. Artificial intelligence-based prediction for cancer susceptibility, recurrence and survival[J]. Progress in Biochemistry and Biophysics, 2022, 49(9): 1687-1702.
[6]	刘富, 梁艺馨, 侯涛, 等. 模糊c-harmonic均值算法在不平衡数据上改进[J]. 吉林大学学报: 工学版, 2021, 51(4): 1447-1453.
	Liu Fu, Liang Yi-xin, Hou Tao, et al. Improvement of fuzzy c-harmonic mean algorithm on unbalanced data[J]. Journal of Jilin University (Engineering and Technology Edition), 2021, 51(4): 1447-1453.
[7]	章鸯, 潘飞燕, 章卫国, 等. 高通量测序在无创产前遗传学诊断中的应用价值[J]. 中国卫生检验杂志, 2022, 32(10): 1249-1253.
	Zhang Yang, Pan Fei-yan, Zhang Wei-guo, et al. Application value of high -throughput sequencing noninvasive prenatal testing in prenatal genetic diagnosis[J]. Chin J Health Lab Tec, 2022, 32(10): 1249-1253.
[8]	方朝剑, 胡新荣. 基于模糊近似度的隐私敏感数据过滤算法[J]. 吉林大学学报: 工学版, 2023, 53(4): 1174-1180.
	Fang Chao-jian, Hu Xin-rong. Privacy-sensitive data filtering algorithm based on fuzzy approximation[J]. Journal of Jilin University (Engineering and Technology Edition), 2023, 53(4): 1174-1180.
[9]	张浩, 李海鹏, 彭国琴, 等. 多层次特征融合表征的图像情感识别[J]. 计算机辅助设计与图形学学报, 2023, 35: 1-11.
	Zhang Hao, Li Hai-peng, Peng Guo-qin, et al. Image emotion recognition via fusion multi-level representations[J]. Journal of Computer-Aided Design & Computer Graphics, 2023, 35: 1-11.
[10]	Alain L U, Melissa S, Péter O, et al. Transcriptome-based identification of novel endotypes in adult atopic dermatitis[J]. Allergy, 2022, 77(5): 1486-1498.
[11]	Sun M, Ding T, Tang X Q, et al. An efficient mixed-model for screening differentially expressed genes of breast cancer based on LR-RF[J]. IEEE/ACM Transactions on Computational Biology and Bioinformatics, 2019, 16(1): 124-130.
[12]	Ball M W, Gorin M A, Drake C G, et al. The landscape of whole-genome alterations and pathologic features in genitourinary malignancies: An analysis of the cancer genome atlas[J]. European Urology Focus, 2017, 3(6): 584-589.
[13]	Bourgade R, Rabilloud N, Perennec T, et al. Deep learning for detecting BRCA mutations in high-grade ovarian cancer based on an innovative tumor segmentation method from whole slide images[J]. Modern Pathology, 2023, 36(11): No.100304.
[14]	Yildirimtepe C F, Ercan C. RGS10 suppression by DNA methylation is associated with low survival rates in colorectal carcinoma[J]. Pathology - Research and Practice, 2022, 236: No.154007.
[15]	Aysegul C, Cenk A A, Yalcin A K. Novel molecular signatures and potential therapeutics in renal cell carcinomas: Insights from a comparative analysis of subtypes[J]. Genomics, 2020, 112(5): 3166-3178.
[16]	Turab N A A, Murtaza R S A, Imtaiyaz H M. Pan-cancer analysis of Chromobox (CBX) genes for prognostic significance and cancer classification[J]. Biochimica et Biophysica Acta (BBA)-Molecular Basis of Disease, 2023, 1869(1): No.166561.
[17]	Goodfellow I J, Jean P A, Mehdi M, et al. Generative adversarial nets[C]∥Proceedings of the 27th International Conference on Neural Information Processing Systems, Montreal, Canada, 2014: 2672-2680.
[18]	Kingma D P. Welling M. Auto-encoding variational Bayes[J/OL].[2023-10-22].
[19]	Precup D, Teh Y W. Wasserstein Generative Adversarial Networks[C]∥Proceedings of the 34th International Conference on Machine Learning, Sydney, Australia, 2017, 70: 214-223.
[20]	Chawla N V, Bowyer K W, Hall L O, et al. SMOTE: synthetic minority over-sampling technique[J]. J. Artif. Int. Res.,2002, 16(1): 321-357.
[21]	周丰丰, 孙燕杰, 范雨思. 基于miRNA组学的数据增强算法[J]. 电子科技大学学报, 2023, 52(2): 182-187.
	Zhou Feng-feng, Sun Yan-jie, Fan Yu-si. Data augmentation algorithm for miRNA omics-based classifications[J]. Journal of University of Electronic Science and Technology of China, 2023, 52(2): 182-187.

Metrics

Viewed

Full text

Abstract

Cited

Shared

Discussed

参数	取值
编码器层数/层	3
编码器网络节点数/个	512， 256， 128
编码器每层激活函数	ReLu， ReLu， ReLu
生成器层数/层	4
生成器网络节点数/个	128， 256， 512， 1 024
生成器激活函数	ReLu， ReLu， ReLu， Tanh
鉴别器层数/层	4
鉴别器网络节点数	512， 256， 128， 1
鉴别器激活函数	Leaky ReLu， Leaky ReLu， Leaky ReLu， Sigmoid

数据集	甲基化特征数	转录组特征数	样本数	正样本数	负样本数	不平衡率Lr（N/P）
ACC	394 014	60 483	76	8	68	8.50
BLCA	382 024	60 483	444	166	278	1.67
BRCA	363 736	60 483	1 112	298	814	2.73
CHOL	378 735	60 483	53	8	45	5.63
COAD	374 805	60 483	459	165	294	1.78
KICH	391 298	60 483	58	8	50	6.25
KIRC	382 720	60 483	274	17	257	15.12
KIRP	383 105	60 483	123	47	76	1.62
LUAD	369 182	60 483	352	7	345	4.24
LUSC	370 619	60 483	639	122	517	5.37
MESO	381 223	60 483	465	73	392	1.13

	特征表征前				ImFeatures
数据集	KNN	SVM	LR	NB	KNN	SVM	LR	NB
ACC	0.816 7	0.683 3	0.816 7	0.633 3	0.883 3	0.883 3	0.950 0	0.933 3
BLCA	0.705 1	0.704 9	0.680 5	0.714 1	0.852 6	0.774 3	0.773 8	0.717 2
BRCA	0.632 9	0.718 0	0.590 4	0.620 5	0.819 1	0.739 3	0.709 1	0.636 6
CHOL	0.616 7	0.666 7	0.800 0	0.933 3	1.000 0	0.933 3	1.000 0	1.000 0
COAD	0.632 9	0.718 0	0.590 4	0.620 5	0.819 1	0.739 3	0.709 1	0.636 6
KICH	1.000 0	1.000 0	1.000 0	0.933 3	1.000 0	1.000 0	1.000 0	1.000 0
KIRC	0.820 0	0.820 0	0.746 7	0.860 0	0.926 7	0.926 7	0.966 7	0.933 3
KIRP	0.871 9	0.787 7	0.882 5	0.787 7	0.977 8	0.893 0	0.935 7	0.829 8
LUAD	0.708 8	0.709 4	0.680 6	0.774 7	0.869 0	0.799 3	0.803 4	0.778 8
LUSC	0.664 8	0.664 8	0.603 2	0.746 4	0.828 5	0.842 3	0.856 3	0.808 3
MESO	0.750 0	0.722 5	0.763 3	0.854 2	0.815 8	0.802 5	0.815 8	0.855 0
数据集	特征表征前				ImFeatures
数据集	KNN	SVM	LR	NB	KNN	SVM	LR	NB
ACC	0.850 0	0.750 0	0.850 0	0.650 0	0.900 0	0.900 0	0.950 0	0.900 0
BLCA	0.705 1	0.704 6	0.680 7	0.714 3	0.852 4	0.774 1	0.773 7	0.717 3
BRCA	0.633 1	0.718 2	0.590 4	0.620 7	0.819 2	0.739 5	0.709 3	0.636 7
CHOL	0.600 0	0.650 0	0.850 0	0.900 0	1.000 0	0.900 0	1.000 0	1.000 0
COAD	0.633 1	0.718 2	0.590 4	0.620 7	0.819 2	0.739 5	0.709 3	0.636 7
KICH	1.000 0	1.000 0	1.000 0	0.950 0	1.000 0	1.000 0	1.000 0	1.000 0
KIRC	0.816 7	0.816 7	0.750 0	0.866 7	0.933 3	0.916 7	0.966 7	0.933 3
KIRP	0.868 9	0.782 2	0.878 9	0.785 6	0.977 8	0.891 1	0.935 6	0.827 8
LUAD	0.708 7	0.710 7	0.682 2	0.776 2	0.870 0	0.800 7	0.804 3	0.780 3
LUSC	0.667 1	0.665 7	0.603 8	0.748 6	0.828 1	0.843 8	0.856 2	0.810 0
MESO	0.755 4	0.728 6	0.764 3	0.858 9	0.819 6	0.807 1	0.816 1	0.857 1

数据名称	准确率	敏感性	特异性	精确率	召回率	马修斯相关系数	ROC曲线下面积
BRCA-Transcriptomics	0.604 0	0.406 3	0.802 2	0.667 1	0.406 3	0.225 1	0.604 2
BRCA-Methylation	0.596 6	0.489 3	0.704 1	0.623 5	0.489 3	0.198 1	0.596 7
BRCA-Multi-omics	0.632 9	0.464 8	0.801 3	0.701 3	0.464 8	0.283 1	0.633 1
LUAD-Transcriptomics	0.612 0	0.464 0	0.760 0	0.657 2	0.464 0	0.235 0	0.612 0
LUAD-Methylation	0.645 4	0.456 2	0.835 2	0.737 9	0.456 2	0.316 5	0.645 7
LUAD-Multi-omics	0.708 8	0.639 0	0.778 3	0.745 3	0.639 0	0.423 4	0.708 7
LUSC-Transcriptomics	0.593 0	0.258 4	0.927 9	0.788 9	0.258 4	0.251 9	0.593 2
LUSC-Methylation	0.576 1	0.318 7	0.836 3	0.557 0	0.318 7	0.147 6	0.577 5
LUSC-Multi-omics	0.664 8	0.375 2	0.959 0	0.920 0	0.375 2	0.415 5	0.667 1

数据集	算法	准确率	敏感性	特异性	精确率	召回率
BRCA	Base	0.632 9	0.464 8	0.801 3	0.701 3	0.464 8
	SMOTE	0.666 7	0.584 7	0.748 3	0.695 6	0.584 7
	ImFeatures	0.813 8	0.815 2	0.812 2	0.813 5	0.815 2
	SMOTE+ImFeatures	0.831 5	0.818 7	0.844 2	0.839 9	0.818 7
LUAD	Base	0.700 6	0.623 0	0.779 3	0.744 7	0.623 0
	SMOTE	0.680 0	0.623 0	0.737 0	0.715 9	0.623 0
	ImFeatures	0.848 3	0.819 0	0.877 0	0.867 6	0.819 0
	SMOTE+ImFeatures	0.828 2	0.802 3	0.853 3	0.853 6	0.802 3
LUSC	Base	0.664 8	0.383 8	0.945 7	0.894 4	0.383 8
	SMOTE	0.602 8	0.316 2	0.889 5	0.770 0	0.316 2
	ImFeatures	0.807 8	0.632 4	0.985 7	0.977 8	0.632 4
	SMOTE+ImFeatures	0.814 7	0.659 0	0.971 4	0.957 8	0.659 0