CGPS: A machine learning-based approach integrating multiple gene set analysis tools for better prioritization of...

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Gene set enrichment (GSE) analyses are vital for interpreting large-scale transcriptome datasets.
• Existing GSE tools have discrepant performances, making optimal result interpretation challenging.
• Current ensemble methods present difficulties for users in selecting the best integrated score.

Purpose of the Study:

• To develop an ensemble method for GSE analysis that integrates results from multiple tools into a single, reliable score.
• To create the first GSE ensemble method leveraging a priori knowledge of pathways and phenotypes.
• To enhance the prioritization of relevant biological pathways from transcriptome data.

Main Methods:

• Developed Combined Gene set analysis incorporating Prioritization and Sensitivity (CGPS), a machine learning-based ensemble method.
• Integrated results from nine prominent GSE tools into a single ensemble score (R score).
• Validated CGPS using 120 simulated and 45 real datasets, comparing it against individual methods and existing ensemble scores.

Main Results:

• The R score from CGPS demonstrated superior prioritization of relevant pathways compared to 10 individual methods and five ensemble scores.
• CGPS successfully identified key cancer-associated pathways, such as the p53 signaling pathway, in panobinostat drug expression data.
• The method effectively captured biological information from diverse gene set collections including KEGG, GO, Reactome, and BioCarta.

Conclusions:

• CGPS provides a robust ensemble approach for gene set enrichment analysis, improving pathway prioritization.
• The method's foundation in a priori knowledge enhances its ability to extract meaningful biological insights.
• CGPS offers a valuable tool for researchers analyzing transcriptome data, with publicly available source code.