Reliable gene signatures for microarray classification: assessment of stability and performance

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Accurate sample classification and identification of gene signatures are crucial for analyzing gene expression data.
• Existing methods can yield unstable gene signatures and unreliable classification accuracy, particularly with sample subsets.
• This can lead to impaired biological conclusions in gene expression studies.

Purpose of the Study:

• To develop a method for robustly classifying samples and identifying stable gene signatures from gene expression data.
• To improve the reliability of classification performance estimates and gene signature stability.
• To provide a framework for selecting optimal classifier and gene selection method combinations for specific datasets.

Main Methods:

• A novel sampling approach was employed, repeatedly evaluating pairwise combinations of gene selection and classification methods on random sample subsets.
• A model scoring system was used to select the most appropriate model for the dataset.
• Consensus gene signatures were constructed by identifying genes consistently selected across multiple samplings.

Main Results:

• The sampling method provided reliable estimates of classification performance, addressing variability observed with subset analysis.
• Stable consensus gene signatures relevant to osteoarthritic cartilage were identified.
• The approach demonstrated superior performance compared to other methods on a breast cancer dataset.

Conclusions:

• The proposed method enhances the reliability of gene expression data analysis by providing stable gene signatures and accurate classification performance estimates.
• This approach is valuable for biological and biomedical applications requiring robust interpretation of gene expression differences.
• The R package is available for public use, facilitating further research in the field.