GO-PCA: An Unsupervised Method to Explore Gene Expression Data Using Prior Knowledge.

Area of Science:

• Genomics
• Bioinformatics
• Systems Biology

Background:

• Genome-wide expression profiling is crucial for analyzing complex biological samples.
• Unsupervised methods like PCA and hierarchical clustering are common but don't leverage gene function knowledge.
• Existing methods struggle to integrate gene correlation with functional relationships.

Purpose of the Study:

• To introduce GO-PCA, an unsupervised method combining PCA and Gene Ontology (GO) enrichment analysis.
• To systematically identify gene sets that are both strongly correlated and functionally related.
• To generate automatically labeled expression signatures for interpretable biological insights.

Main Methods:

• GO-PCA combines principal component analysis (PCA) with nonparametric Gene Ontology (GO) enrichment analysis.
• Identifies functionally related gene sets within expression data.
• Generates interpretable expression signatures with functional labels.

Main Results:

• Applied to human and mouse hematopoietic cell data, GO-PCA identified key lineages.
• Analysis of human glioblastoma (GBM) data revealed signatures for four out of five subtypes.
• Demonstrates GO-PCA's ability to reduce large gene expression matrices into smaller, interpretable signatures.

Conclusions:

• GO-PCA is a powerful and versatile exploratory method for gene expression data.
• Facilitates hypothesis generation, further analysis design, and cross-dataset functional comparisons.
• Provides readily interpretable representations of biological similarities and differences.