Supervised Discriminative Sparse PCA for Com-Characteristic Gene Selection and Tumor Classification on Multiview Biological Data

Chun-Mei Feng, Yong Xu, Jin-Xing Liu, Ying-Lian Gao, Chun-Hou Zheng

Abstract

Principal Component Analysis (PCA) has been used to study the pathogenesis of diseases. To enhance the interpretability of classical PCA, various improved PCA methods have been proposed to date. Among these, a typical method is the so-called sparse PCA, which focuses on seeking sparse loadings. However, the performance of these methods is still far from satisfactory due to their limitation of using unsupervised learning methods; moreover, the class ambiguity within the sample is high. To overcome this problem, this study developed a new PCA method, which is named the Supervised Discriminative Sparse PCA (SDSPCA). The main innovation of this method is the incorporation of discriminative information and sparsity into the PCA model. Specifically, in contrast to the traditional sparse PCA, which imposes sparsity on the loadings, here, sparse components are obtained to represent the data. Furthermore, via linear transformation, the sparse components approximate the given label information. On the one hand, sparse components improve interpretability over traditional PCA, while on the other hand, they are have discriminative abilities suitable for classification purposes. A simple algorithm is developed and its convergence proof is provided. The SDSPCA has been applied to common characteristic gene selection (com-characteristic gene) and tumor classification on multi-view biological data. The sparsity and classification performance of the SDSPCA are empirically verified via abundant, reasonable, and effective experiments, and the obtained results demonstrate that SDSPCA outperforms other state-of-the-art methods.

Introduction

In the field of genomics, researchers frequently encounter the "curse of dimensionality," where datasets contain expression levels for over 20,000 genes but only a few hundred samples. Standard Principal Component Analysis (PCA) is a common tool for simplifying such data, yet it possesses two significant drawbacks for clinical application: it is unsupervised, meaning it may ignore variance critical for disease classification, and it produces "dense" results where every gene contributes to the model, hindering biological interpretability.

To address these issues, this research introduces Supervised Discriminative Sparse PCA (SDSPCA). This method combines the interpretability of sparse PCA with the predictive power of supervised learning, specifically designed to identify "com-characteristic" genes—markers consistently important across multiple cancer types.

Methodology

SDSPCA enhances the traditional PCA objective function by integrating two critical constraints:

  1. Discriminative Supervision: By incorporating a class indicator matrix directly into the optimization process, the model forces principal components to actively approximate sample labels rather than just maximizing variance.
  2. $L_{2,1}$-Norm Sparsity: Unlike standard sparse PCA methods that sparsify loadings, SDSPCA imposes sparsity on the components themselves. This effectively serves as a feature selection mechanism, "turning off" irrelevant genes and producing a concise list of markers for biological study.

The model was evaluated using high-dimensional gene expression data from The Cancer Genome Atlas (TCGA), specifically focusing on a multiview dataset of 20,502 genes across 643 samples, categorized into four classes: Pancreatic cancer (PAAD), Head and Neck squamous cell carcinoma (HNSC), Cholangiocarcinoma (CHOL), and normal tissue.

Results and Discussion

Classification Performance and Robustness

SDSPCA achieved an average classification accuracy of 91.82%, significantly outperforming classical PCA (88.25%) and other sparse variants like SPCArt (90.85%). A key advantage of the model is its stability; sensitivity analysis demonstrates that the classification accuracy remains high and consistent across a wide range of the hyperparameters $\alpha$ and $\beta$.

Figure 1: Sensitivity analysis showing the stability of classification accuracy (ACC) across different values of hyperparameters alpha and beta.

Gene Selection and Biological Validation

By treating sparse components as "codes," SDSPCA provides superior interpretability. It successfully identified 60 com-characteristic genes, including critical markers like PDGFRB and ABCC1. These genes are known in medical literature to be vital for cancer progression but are often overlooked by unsupervised methods.

Figure 6: Comparison of com-characteristic genes identified by SDSPCA versus other methods, highlighting unique markers like PDGFRB and ABCC1.

Further biological validation using KEGG pathway analysis revealed that the selected genes are heavily involved in pathways such as "Focal adhesion" and "ECM-receptor interaction," which are fundamental to cancer metastasis.

Conclusion

SDSPCA represents a significant advancement in dimensionality reduction for biological data. By injecting label information into the PCA process and enforcing component sparsity, it reduces class ambiguity and identifies robust cross-disease markers. While currently a linear model, its computational efficiency and high accuracy make it a powerful tool for large-scale genomic analysis. Future work will focus on extending the model to capture non-linear structures and improving its robustness to outliers.

Summary updated

Abstract

Principal Component Analysis (PCA) has been used to study the pathogenesis of diseases. To enhance the interpretability of classical PCA, various improved PCA methods have been proposed to date. Among these, a typical method is the so-called sparse PCA, which focuses on seeking sparse loadings. However, the performance of these methods is still far from satisfactory due to their limitation of using unsupervised learning methods; moreover, the class ambiguity within the sample is high. To overcome this problem, this study developed a new PCA method, which is named the Supervised Discriminative Sparse PCA (SDSPCA). The main innovation of this method is the incorporation of discriminative information and sparsity into the PCA model. Specifically, in contrast to the traditional sparse PCA, which imposes sparsity on the loadings, here, sparse components are obtained to represent the data. Furthermore, via linear transformation, the sparse components approximate the given label information. On the one hand, sparse components improve interpretability over traditional PCA, while on the other hand, they are have discriminative abilities suitable for classification purposes. A simple algorithm is developed and its convergence proof is provided. The SDSPCA has been applied to common characteristic gene selection (com-characteristic gene) and tumor classification on multi-view biological data. The sparsity and classification performance of the SDSPCA are empirically verified via abundant, reasonable, and effective experiments, and the obtained results demonstrate that SDSPCA outperforms other state-of-the-art methods.
BibTeX
@article{feng2019supervised,
  author = {Chun-Mei Feng and Yong Xu and Jin-Xing Liu and Ying-Lian Gao and Chun-Hou Zheng},
  title = {{Supervised Discriminative Sparse PCA for Com-Characteristic Gene Selection and Tumor Classification on Multiview Biological Data}},
  journal = {IEEE Trans. Neural Networks Learn. Syst.},
  year = {2019},
  url = {https://openreview.net/forum?id=-bjDfmQL1V},
}
Figure