Selecting differentially expressed genes using minimum probability of classification error.

Area of Science:

• Bioinformatics and Computational Biology
• Cancer Genomics
• Biomarker Discovery

Background:

• Identifying differentially expressed genes is crucial for diagnosing diseases.
• Accurate classification requires effective genetic markers.
• Current methods may not optimally rank genes for diagnostic potential.

Purpose of the Study:

• To develop and evaluate a novel gene ranking method based on Minimum Probability of Classification Errors (MPE).
• To identify a minimal set of genetic markers for disease diagnosis.
• To compare the performance of MPE-ranked genes against p-value-based ranking.

Main Methods:

• Utilized a Bayesian decision-making algorithm to compute MPE for each gene.
• Employed quantile-based probability density estimation for gene probability distributions.
• Validated marker performance using Support Vector Machine (SVM) and modified Naive Bayes classifiers on colon cancer, leukemia, and hereditary breast cancer datasets.

Main Results:

• Achieved 96.77% accuracy for colon cancer and 97.06% for leukemia using only five genes.
• Attained 100% accuracy for hereditary breast cancer with just three genes.
• MPE-ranked genes demonstrated superior or equal performance compared to p-value-ranked genes.

Conclusions:

• The Minimum Probability of Classification Errors (MPE) method is effective for identifying potent genetic diagnostic markers.
• This approach offers a promising strategy for biomarker discovery in complex diseases.
• MPE-based gene selection provides a robust alternative to traditional p-value ranking for diagnostic applications.