Tumor classification by combining PNN classifier ensemble with neighborhood rough set based gene reduction.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Gene expression profiles (GEP) are crucial for molecular tumor classification.
• High-dimensional and small-sample tumor datasets pose significant challenges for accurate classification.
• Existing GEP-based methods struggle with data complexity.

Purpose of the Study:

• To develop a robust tumor classification approach addressing high-dimension, small-sample challenges.
• To enhance the accuracy and reliability of molecular tumor subtype diagnosis.
• To identify functionally relevant genes for carcinogenesis.

Main Methods:

• Utilized an ensemble of probabilistic neural networks (PNN) combined with neighborhood rough set model-based gene reduction.
• Employed an iterative search margin algorithm for initial informative gene selection.
• Refined gene subsets using gene reduction and integrated base PNN classifiers via majority voting.

Main Results:

• Achieved high and stable classification performance on tumor datasets.
• Demonstrated robustness to the number of initially selected genes.
• Showcased competitive performance compared to existing GEP-based methods.

Conclusions:

• The proposed method offers a convincing and reliable approach for GEP-based tumor classification.
• Selected gene subsets are biologically relevant to carcinogenesis, enabling cross-verification.
• This method effectively overcomes the limitations of high-dimensional, small-sample tumor data.