Parallel gene selection and dynamic ensemble pruning based on Affinity Propagation.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Gene selection and sample classification from gene expression data are crucial in bioinformatics.
• High dimensionality and small sample sizes in microarray data present challenges for accurate gene selection.
• Extended rough sets effectively handle redundancy and numerical data but are computationally intensive.

Purpose of the Study:

• To accelerate the computationally demanding approximation calculations in neighborhood rough set-based gene selection.
• To develop a dynamic ensemble pruning approach for reducing memory and computational costs.
• To improve classification performance in gene expression data analysis.

Main Methods:

• A parallel computation method was proposed to expedite the calculation of approximations for the intersection neighborhood rough set.
• A novel dynamic ensemble pruning strategy utilizing Affinity Propagation clustering was introduced.
• The proposed methods were evaluated on Arabidopsis thaliana stress response datasets.

Main Results:

• The parallel computation significantly reduces the time required for approximation calculations.
• The dynamic ensemble pruning approach effectively lowers memory usage and computational demands.
• Experimental results demonstrated superior classification performance compared to ensemble methods with pre-selected genes.

Conclusions:

• The proposed parallel and dynamic pruning methods offer an efficient solution for gene selection and sample classification.
• This approach enhances the practical applicability of rough set methods in bioinformatics.
• The study highlights the potential for improved accuracy and efficiency in analyzing complex biological datasets.