Comparison and evaluation of pathway-level aggregation methods of gene expression data

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Microarray experiments generate large-scale gene expression data.
• Pathway-level analysis offers functional insights by aggregating gene expression data.
• Existing pathway aggregation methods lack comprehensive comparative evaluation.

Purpose of the Study:

• To comprehensively evaluate and compare six prominent gene expression data aggregation methods at the pathway level.
• To identify the most effective methods for pathway-level analysis in bioinformatics.

Main Methods:

• Compared six pathway-level aggregation methods: Mean all, Mean CORGs, ASSESS, PCA, PLS, and Mean top 50%.
• Utilized seven independent datasets with various phenotypes for stringent benchmarking.
• Assessed performance using classification accuracy and pathway signature correlation across KEGG pathways.

Main Results:

• ASSESS and Mean top 50% achieved the highest classification accuracy and correlation.
• Mean all exhibited the lowest accuracy.
• Mean CORGs and PLS showed the greatest discordance in pathway signature correlation.

Conclusions:

• ASSESS and Mean top 50% are recommended as preferred methods for pathway-level aggregation.
• The study highlights the need for developing novel, improved pathway aggregation methods.