Kernel-based hierarchical structural component models for pathway analysis on survival phenotype

Area of Science:

• Genomics and Bioinformatics
• Cancer Research
• Computational Biology

Background:

• High-throughput sequencing, including RNA-sequencing (RNA-seq), has revolutionized gene expression analysis.
• Traditional pathway analysis methods often overlook inter-pathway correlations and overlapping biomarkers.
• Existing approaches typically assume linear gene effects on phenotypes, limiting their scope.

Purpose of the Study:

• To develop the HisCoM-KernelS model for identifying survival phenotype-related pathways.
• To accommodate complex, nonlinear relationships between genes and survival outcomes.
• To account for inter-pathway correlations in pathway-based survival analysis.

Main Methods:

• Applied the HisCoM-KernelS model to the TCGA pancreatic ductal adenocarcinoma (PDAC) RNA-seq dataset.
• Utilized kernel machine regression to model pathway effects on survival, incorporating gene-pathway structures.
• Estimated model parameters via alternating least squares and assessed pathway significance using permutation tests.

Main Results:

• HisCoM-KernelS identified significant pathways associated with pancreatic cancer survival.
• The model demonstrated a superior balance of detection rate and significant pathways compared to HisCoM-PAGE, Global Test, GSEA, and CoxKM.
• Gaussian kernel integration in HisCoM-KernelS enhanced performance.

Conclusions:

• HisCoM-KernelS effectively extends pathway analysis to survival outcomes by capturing nonlinear gene effects and inter-pathway correlations.
• The model's application to TCGA PDAC data highlights its utility in identifying biologically relevant pathways.
• HisCoM-KernelS provides a robust tool for survival phenotype research using high-throughput sequencing data.