Factors influencing the statistical power of complex data analysis protocols for molecular signature development from...

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Statistical significance and reproducibility are crucial for molecular signatures from high-throughput data.
• Current methods often rely on univariate tests, with limited understanding of multivariate analysis power.

Purpose of the Study:

• To investigate factors influencing the statistical power of multivariate analysis protocols for molecular signature development.
• To identify key components of analysis that impact power and reproducibility.

Main Methods:

• Empirical analysis of 7 large microarray cancer outcome prediction datasets.
• Supplementary simulations to assess the effects of different analysis components.
• Comparison with prior analyses of the same datasets.

Main Results:

• Specific choices in error metric, classifier, error estimator, and event balancing significantly affect statistical power.
• These effects are compounding, demonstrating a substantial impact on results.
• Prior analyses showing poor prediction accuracy were likely due to under-powered methods.

Conclusions:

• Optimizing multivariate analysis protocols can reduce sample size needs for high-throughput studies.
• Identified factors allow for more efficient and powerful molecular signature development.
• Revises previous conclusions about the limitations of microarray-based outcome prediction.