Enhancing Characteristic Gene Selection and Tumor Classification by the Robust Laplacian Supervised Discriminative...

Area of Science:

• Genomics and Bioinformatics
• Machine Learning in Biology
• Computational Biology

Background:

• Gene expression data analysis is crucial in genomics.
• High dimensionality and small sample sizes necessitate dimensionality reduction.
• Existing methods like PCA and sparse PCA have limitations in robustness and capturing data structure.

Purpose of the Study:

• To propose a novel PCA-based method, RLSDSPCA, for improved gene expression data analysis.
• To address limitations of existing methods by incorporating robustness, label information, sparsity, and geometrical structures.
• To enhance characteristic gene selection and tumor classification.

Main Methods:

• Developed Robust Laplacian Supervised Discriminative Sparse PCA (RLSDSPCA).
• Enforced L2,1 norm on the error function for robustness.
• Incorporated graph Laplacian into supervised discriminative sparse PCA.
• Applied RLSDSPCA to gene expression data for characteristic gene selection and tumor classification.

Main Results:

• RLSDSPCA effectively identified new pathogenic genes associated with diseases.
• The method demonstrated superior performance in tumor classification compared to state-of-the-art techniques.
• RLSDSPCA achieved the best performance across major metrics in tumor classification tasks.

Conclusions:

• RLSDSPCA is an effective method for characteristic gene selection and tumor classification.
• The proposed method offers improved robustness and better capture of data structures.
• RLSDSPCA shows significant potential for advancing genomic research and disease understanding.