DSTPCA: Double-Sparse Constrained Tensor Principal Component Analysis Method for Feature Selection.

Area of Science:

• Bioinformatics and Computational Biology
• Genomics and Gene Expression Analysis
• Data Mining and Machine Learning

Background:

• Identifying differentially expressed genes is crucial in biological research.
• Traditional Principal Component Analysis (PCA) methods overlook data's spatial geometric structure.
• Tensor robust principal component analysis (TRPCA) preserves spatial structure using tensor decomposition.

Purpose of the Study:

• To introduce an improved feature selection method based on TRPCA.
• To enhance gene selection accuracy by incorporating spatial structure and sparsity.
• To develop a robust method for filtering redundant genes and identifying disease-associated genes.

Main Methods:

• Developed Double-Sparse Constrained Tensor Principal Component Analysis (DSTPCA) by introducing L_2,1-norm regularization.
• Applied double sparse constraints to the objective function for noise reduction and sparse results.
• Utilized the alternating direction method of multipliers (ADMM) algorithm for optimal problem solving.

Main Results:

• DSTPCA effectively filtered out redundant genes while screening for disease-associated genes.
• The method demonstrated superior performance compared to existing methods across various datasets.
• Preservation of spatial geometric structure in high-dimensional gene expression data was achieved.

Conclusions:

• DSTPCA offers a significant advancement in gene feature selection for bioinformatics.
• The method's ability to handle spatial structures and enforce sparsity improves biological insights.
• DSTPCA is a promising tool for identifying key genes in complex diseases.