Measure of Strength of Evidence for Visually Observed Differences between Subpopulations.

Area of Science:

• Biostatistics
• Computational Biology
• Genomics

Background:

• Assessing subpopulation differences is crucial in high-dimensional data.
• Conventional statistical methods face challenges in high-dimensional and high-signal contexts.
• Existing permutation tests can be suboptimal for pairwise comparisons.

Purpose of the Study:

• To introduce the Population Difference Criterion for measuring statistical significance of visually-observed subpopulation differences.
• To address limitations of existing methods in high-dimensional and high-signal data.
• To enhance the power and reliability of subpopulation comparisons.

Main Methods:

• Development of the Population Difference Criterion.
• Analysis of balanced permutation approaches for improved statistical power.
• Implementation of bootstrap confidence intervals for quantifying permutation uncertainty.
• Application to modern cancer subpopulation data.

Main Results:

• The Population Difference Criterion provides a robust measure of subpopulation difference significance.
• Balanced permutation tests demonstrate superior power in high-signal contexts compared to conventional methods.
• Bootstrap confidence intervals effectively quantify uncertainty arising from permutation variation.
• The proposed methods show practical utility in analyzing cancer subpopulation data.

Conclusions:

• The Population Difference Criterion offers a valuable tool for statistical analysis of subpopulation differences.
• Balanced permutations and bootstrap confidence intervals enhance the reliability of findings in complex datasets.
• The approach is effective for real-world applications, such as cancer genomics.