Unimodal transform of variables selected by interval segmentation purity for classification tree modeling of...

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Classification and Regression Trees (CART) are prone to overfitting when modeling microarray gene expression data.
• Identifying significant genes is crucial for mitigating overfitting, but existing methods struggle with multi-modal expression patterns.
• Complex gene expression patterns can lead to underfitting or overfitting in CART models.

Purpose of the Study:

• To propose a novel strategy, unimodal transform of variables selected by interval segmentation purity (UTISP), to enhance CART modeling for microarray data.
• To address the limitations of existing gene identification methods in capturing complex, multi-modal expression patterns.
• To improve the performance of CART in combating overfitting and underfitting in microarray data analysis.

Main Methods:

• A variable selection method based on interval segmentation purity is used to identify significant genes with varied expression patterns.
• Unimodal transform is applied for feature extraction, creating unimodal featured variables.
• The proposed UTISP strategy is integrated with CART for modeling microarray gene expression data.

Main Results:

• The UTISP-based CART strategy effectively identifies significant genes with complex expression patterns.
• Unimodal feature extraction improves the input data for CART modeling.
• The developed strategy demonstrates superior performance compared to k-nearest neighbors and other CART-based gene identification methods in two microarray datasets.

Conclusions:

• The UTISP-based CART strategy offers a promising approach for more accurate microarray data analysis.
• This method enhances CART's ability to handle complex gene expression patterns, reducing overfitting and underfitting.
• UTISP-based CART shows significant potential for improving the reliability of gene expression data modeling.