A novel dimension reduction algorithm based on weighted kernel principal analysis for gene expression data.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Gene expression datasets are high-dimensional with small sample sizes, often containing irrelevant genes.
• Directly applying machine learning to such data is computationally expensive and can hinder classification accuracy.

Purpose of the Study:

• To propose a novel dimension reduction algorithm, weighted kernel principal component analysis (WKPCA), for gene expression data.
• To enhance the efficiency and effectiveness of machine learning models in classifying diseases based on gene expression.

Main Methods:

• Developed a dimension-reduction algorithm based on weighted kernel principal component analysis (WKPCA).
• Constructed kernel function weights using kernel matrix eigenvalues and combined multiple kernel functions.
• Introduced t-class kernel functions for improved WKPCA efficiency, with theoretical proofs.
• Utilized Naive Bayes, K-nearest neighbour, random forest, iterative random forest, and support vector machine classifiers on 6 real gene expression datasets.

Main Results:

• WKPCA significantly improved the classification performance of five machine learning algorithms.
• Demonstrated superior performance compared to all-variable models, linear principal component analysis, and single-kernel methods.
• The cumulative optimal performance rate effectively measured the combined performance of WKPCA and machine learning.

Conclusions:

• WKPCA is an effective dimension reduction technique for high-dimensional gene expression data.
• The proposed method enhances the accuracy and efficiency of machine learning-based disease classification.
• WKPCA offers a valuable approach for analyzing complex genomic datasets.