Learning Differentially Expressed Gene Pairs in Microarray Data.

Area of Science:

• Bioinformatics
• Computational Biology
• Gene Expression Analysis

Background:

• Existing methods for identifying differentially expressed genes (DEGs) often rank genes individually, potentially missing important gene combinations.
• This individual ranking approach may overlook genes with subtle individual but significant combined discriminatory power.

Purpose of the Study:

• To propose a novel impurity metric for DEG identification that considers feature split intervals.
• To evaluate the proposed method's ability to identify significant gene combinations in microarray data analysis.

Main Methods:

• Developed a new impurity metric optimizing the number of split intervals for maximal discrimination.
• Evaluated the method on a synthesized noisy rectangular grid dataset to recognize feature pairs.
• Applied the method to colon microarray data for DEG identification and prescreening.

Main Results:

• Successfully recognized a significant feature pair forming a rectangular grid pattern in synthesized data.
• Demonstrated the proposed method as a viable alternative to Fisher's test for gene prescreening.
• Showcased improved performance of the Support Vector Machine-Recursive Feature Elimination (SVM-RFE) method when using the proposed prescreening.

Conclusions:

• The proposed impurity metric effectively identifies gene combinations with high discriminatory power.
• This method offers an improved approach for DEG identification and gene prescreening in microarray analysis.
• The novel metric enhances the performance of subsequent machine learning-based gene selection techniques.