Integrating biological knowledge and gene expression data using pathway-guided random forests: a benchmarking study.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• High-throughput technologies generate comprehensive omics data for individual characterization.
• Predicting disease status from omics data using computational models is challenging.
• Integrating external biological pathway knowledge can enhance predictive modeling.

Purpose of the Study:

• To compare four random forest-based methods for omics data integration.
• To evaluate the performance of different pathway-guided random forest approaches.
• To provide guidance on selecting appropriate methods based on pathway relevance.

Main Methods:

• Utilized two simulation studies and nine experimental omics datasets.
• Compared four published random forest-based computational approaches.
• Assessed methods based on their ability to predict disease status.

Main Results:

• The self-sufficient prediction error approach is recommended when many relevant pathways are expected.
• Hunting and learner of functional enrichment methods are suitable for scenarios with few relevant pathways.
• The hybrid approach (synthetic features) demonstrated a high false discovery rate and is not recommended.

Conclusions:

• The choice of pathway integration method impacts the accuracy of omics-based disease prediction.
• Specific methods are better suited for different pathway landscape scenarios.
• Software and datasets are publicly available for reproducibility and further research.