Improvement of variables interpretability in kernel PCA

被引:4
作者
Briscik, Mitja [1 ]
Dillies, Marie-Agnes [2 ]
Dejean, Sebastien [1 ]
机构
[1] Univ Toulouse, Inst Math Toulouse, UPS, CNRS,UMR5219, F-31062 Toulouse 9, France
[2] Univ Paris Cite, Inst Pasteur, Bioinformat & Biostat Hub, F-75015 Paris, France
关键词
Kernel PCA; Relevant variables; Unsupervised learning; Kernel methods; RACGAP1; EXPRESSION; MAJOR DETERMINANT; ASPM GENE; ADAM10; CANCER; IDENTIFICATION; METASTASIS; BIOMARKER; TARGET; CCNB1;
D O I
10.1186/s12859-023-05404-y
中图分类号
Q5 [生物化学];
学科分类号
071010 ; 081704 ;
摘要
BackgroundKernel methods have been proven to be a powerful tool for the integration and analysis of high-throughput technologies generated data. Kernels offer a nonlinear version of any linear algorithm solely based on dot products. The kernelized version of principal component analysis is a valid nonlinear alternative to tackle the nonlinearity of biological sample spaces. This paper proposes a novel methodology to obtain a data-driven feature importance based on the kernel PCA representation of the data.ResultsThe proposed method, kernel PCA Interpretable Gradient (KPCA-IG), provides a data-driven feature importance that is computationally fast and based solely on linear algebra calculations. It has been compared with existing methods on three benchmark datasets. The accuracy obtained using KPCA-IG selected features is equal to or greater than the other methods' average. Also, the computational complexity required demonstrates the high efficiency of the method. An exhaustive literature search has been conducted on the selected genes from a publicly available Hepatocellular carcinoma dataset to validate the retained features from a biological point of view. The results once again remark on the appropriateness of the computed ranking.ConclusionsThe black-box nature of kernel PCA needs new methods to interpret the original features. Our proposed methodology KPCA-IG proved to be a valid alternative to select influential variables in high-dimensional high-throughput datasets, potentially unravelling new biological and medical biomarkers.
引用
收藏
页数:21
相关论文
共 78 条
[61]  
Scholkopf B., 1997, Artificial Neural Networks - ICANN '97. 7th International Conference Proceedings, P583, DOI 10.1007/BFb0020217
[62]  
Scholkopf B, 2003, KERNEL METHODS COMPU
[63]  
Scholkopf B, 2002, Learning with Kernels: Support Vector Machines, Regularization, Optimization, and Beyond, DOI [DOI 10.7551/MITPRESS/4175.001.0001, 10.7551/ mitpress/4175.001.0001]
[64]  
Simonyan K, 2014, Arxiv, DOI [arXiv:1312.6034, DOI 10.48550/ARXIV.1312.6034]
[65]  
Smilkov D, 2017, Arxiv, DOI arXiv:1706.03825
[66]   Targeting cyclin-dependent kinase 1 (CDK1) in cancer: molecular docking and dynamic simulations of potential CDK1 inhibitors [J].
Sofi, Shazia ;
Mehraj, Umar ;
Qayoom, Hina ;
Aisha, Shariqa ;
Almilaibary, Abdullah ;
Alkhanani, Mustfa ;
Mir, Manzoor Ahmad .
MEDICAL ONCOLOGY, 2022, 39 (09)
[67]  
Sundararajan M, 2017, Arxiv, DOI [arXiv:1703.01365, 10.48550/arXiv.1703.01365]
[68]  
Springenberg JT, 2015, Arxiv, DOI [arXiv:1412.6806, 10.48550/arXiv.1412.6806, DOI 10.48550/ARXIV.1412.6806]
[69]  
Uurtio V, 2019, PR MACH LEARN RES, V97
[70]   Increased expression of RRM2 by human papillomavirus E7 oncoprotein promotes angiogenesis in cervical cancer [J].
Wang, N. ;
Zhan, T. ;
Ke, T. ;
Huang, X. ;
Ke, D. ;
Wang, Q. ;
Li, H. .
BRITISH JOURNAL OF CANCER, 2014, 110 (04) :1034-1044