An adaptive feature selection algorithm based on MDS with uncorrelated constraints for tumor gene data...

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• DNA microarray technology enables gene-level cancer studies.
• Current unsupervised feature selection methods overlook gene correlations, leading to redundant gene selection and reduced clustering accuracy.

Purpose of the Study:

• To propose an adaptive feature selection model that addresses the limitations of existing methods in gene expression data analysis.
• To improve clustering performance in cancer gene studies by selecting more relevant and non-redundant genes.

Main Methods:

• Developed an adaptive feature selection model incorporating a reconstructed coefficient matrix with constraints.
• Transformed high-dimensional gene data into a low-dimensional space to prevent over-focusing on similar attributes.
• Utilized Alternative Optimization (AO) to manage non-convex optimization challenges.

Main Results:

• The proposed model demonstrated superior performance on four cancer datasets.
• Achieved higher clustering accuracy compared to existing models.
• Resulted in a sparser selection of relevant genes.

Conclusions:

• The proposed adaptive feature selection model effectively enhances gene selection for cancer studies.
• Considering gene correlations and utilizing dimensionality reduction improves clustering outcomes.
• The model offers a promising approach for analyzing complex genomic data in cancer research.