Pathway hunting by random survival forests.

Area of Science:

• Genomics
• Bioinformatics
• Statistical Genetics

Background:

• Pathway and gene set analysis are common in genomics.
• Current methods often overlook marker correlations and interactions.
• Existing random forest approaches analyze pathways independently, missing cross-talk.

Purpose of the Study:

• To develop a new pathway hunting algorithm for survival outcomes.
• To incorporate gene correlation and genomic interactions into pathway prioritization.
• To address the limitation of separate pathway analysis in previous methods.

Main Methods:

• Utilized random survival forests for pathway analysis.
• Developed a novel algorithm to prioritize pathways based on gene correlation and interactions.
• Compared the proposed method against five popular pathway testing techniques.

Main Results:

• The proposed method demonstrated superior performance compared to five existing pathway testing methods.
• Favorable results were observed using both synthetic and real genomic data.
• The methodology proved efficient and powerful for high-dimensional genomic data.

Conclusions:

• The developed algorithm offers an effective framework for pathway modeling in high-dimensional genomic data.
• The approach successfully accounts for complex gene correlations and interactions.
• This method enhances the ability to identify significant pathways in survival outcome studies.