Identifying significantly impacted pathways: a comprehensive review and assessment.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• High-throughput experiments frequently compare phenotypes (e.g., disease vs. healthy) to understand biological mechanisms.
• Over 70 pathway analysis methods exist, categorized as non-topology-based (non-TB) and topology-based (TB).
• Existing reviews lack a systematic, large-scale assessment, and many methods assume uniform p-values under the null hypothesis, which is often inaccurate.

Purpose of the Study:

• To conduct the most comprehensive comparative study of pathway analysis methods to date.
• To evaluate the performance and bias of 13 widely used pathway analysis methods.
• To establish a reliable benchmark for future pathway analysis method development and testing.

Main Methods:

• Compared the performance of 13 pathway analysis methods across 1085 analyses.
• Utilized 2601 samples from 75 human disease datasets and 121 samples from 11 mouse knockout datasets.
• Investigated the bias of each method under the null hypothesis.

Main Results:

• No single pathway analysis method proved perfect.
• Topology-based (TB) methods generally demonstrated superior performance compared to non-topology-based (non-TB) methods.
• Most evaluated methods exhibited bias, potentially leading to skewed results under the null hypothesis.

Conclusions:

• Topology-based methods are generally more effective due to their consideration of pathway structure.
• The identified biases highlight limitations in current pathway analysis approaches.
• The study provides a crucial benchmark for assessing and improving future pathway analysis tools.