Do two machine-learning based prognostic signatures for breast cancer capture the same biological processes?

Area of Science:

• Oncology
• Bioinformatics
• Genomics

Background:

• Prognostic signatures for early breast cancer often lack gene overlap.
• Previous explanations cited machine-learning limitations, questioning the biological relevance of gene lists.
• Claims suggest different signatures capture similar biological pathways.

Purpose of the Study:

• To rigorously evaluate the claim that distinct prognostic gene lists for breast cancer identify similar biological processes.
• To scrutinize the concordance between two prominent gene lists used in large-scale validation efforts.

Main Methods:

• Performed enrichment analysis on two key breast cancer prognostic gene lists.
• Employed statistical methods to control for multiple testing, accounting for gene ontology structure.
• Assessed the biological process and pathway concordance between the gene signatures.

Main Results:

• Found statistically significant concordance only for cell proliferation, a known prognostic factor.
• Contradicted previous publications claiming wider biological process overlap.
• Identified that prior claims often lacked statistical rigor or addressed different research questions.

Conclusions:

• The biological processes captured by different early breast cancer prognostic signatures show limited concordance beyond cell proliferation.
• Claims of broader pathway overlap are often not statistically supported or misinterpret the research question.
• Cell proliferation remains the only robustly identified shared biological process.