Illustrations on Using the Distribution of a P-value in High Dimensional Data Analyses

Area of Science:

• Statistics
• Bioinformatics
• Genomics

Background:

• High-dimensional experiments generate large datasets requiring robust statistical analysis.
• Methods utilizing P-value distributions are increasingly common for analyzing such data.
• The validity of these methods hinges on accurate P-value computation and independence.

Purpose of the Study:

• To investigate the impact of invalid P-values and P-value correlations on statistical analyses.
• To examine the distribution of P-values under null and alternative hypotheses.
• To provide insights into potential biases in high-dimensional data analysis.

Main Methods:

• Analysis of P-value distributions derived from test statistics.
• Focus on the number of rejected null hypotheses as a test statistic.
• Demonstration using simulated and real-world microarray data.

Main Results:

• Invalid P-value computations can introduce significant effects into P-value distributions.
• Correlated P-values can also lead to misleading results, even with correct distributions.
• Illustrates the potential for bias in high-dimensional statistical inference.

Conclusions:

• Emphasizes the critical importance of valid P-value generation in statistical analysis.
• Highlights the risks associated with P-value correlation in high-dimensional settings.
• Suggests careful consideration of P-value properties for reliable data interpretation.