Tumor classification and marker gene prediction by feature selection and fuzzy c-means clustering using microarray...

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Development of novel computational models for tumor classification and gene prediction using DNA microarrays.
• Utilizes Self-Organizing Maps (SOMs) for summarizing gene expression profiles.
• Employs Fuzzy C-means clustering for tumor sample classification.

Purpose of the Study:

• To develop and validate novel models for accurate tumor classification and prediction of marker genes from DNA microarray data.
• To assess the performance of the proposed models against existing methods.
• To highlight the importance of feature selection in microarray data analysis.

Main Methods:

• Gene expression data summarization using optimally selected Self-Organizing Maps (SOMs).
• Tumor sample classification via Fuzzy C-means clustering.
• Marker gene prediction using manual (SOM component plane visualization) or automatic (Fisher's linear discriminant) feature selection.

Main Results:

• Models tested on four diverse datasets: Leukemia, Colon cancer, Brain tumors, and NCI cancer cell lines.
• Achieved significantly reduced error rates in class prediction compared to other approaches.
• Demonstrated the critical role of feature selection in enhancing microarray data analysis.

Conclusions:

• The developed models effectively identify marker genes with high predictive potential, outperforming existing methods.
• Models show promise for medical diagnostics and offer insights into cancer classification.
• Identified inherent limitations in tumor classification from microarray data related to data class size and internal class structure.