The L1-version of the Cramér-von Mises test for two-sample comparisons in microarray data analysis

Area of Science:

• Biostatistics
• Computational Biology
• Genomics

Background:

• Distribution-free statistical tests are crucial for accurate p-value computation in multiple testing procedures, particularly in gene expression analysis.
• The Cramér-von Mises (L(2)) test is a robust distribution-free option but computationally intensive for large datasets.
• Efficient algorithms are needed to compute exact quantiles for these tests in finite samples.

Purpose of the Study:

• To introduce and evaluate an efficient algorithm for computing exact quantiles of an L(1)-distance test statistic.
• To compare the performance and statistical power of the L(1)-distance test against the Cramér-von Mises (L(2)) test and other classical methods.
• To assess the utility of the L(1)-distance test for analyzing microarray data, specifically in childhood leukemia studies.

Main Methods:

• Development of an efficient numerical algorithm for calculating exact quantiles of the L(1)-distance test statistic.
• Comparative analysis of the L(1)-distance test with the Cramér-von Mises (L(2)) test and two other classical tests.
• Validation using both simulated datasets and a comprehensive set of real-world microarray data from childhood leukemia patients.

Main Results:

• The L(1)-distance test demonstrates comparable statistical power to the Cramér-von Mises (L(2)) test.
• The developed algorithm significantly reduces the computational time and space requirements compared to the L(2) test.
• Exact null distribution quantiles can be computed for larger sample sizes using the L(1)-distance test.

Conclusions:

• The L(1)-distance test is a computationally efficient and powerful distribution-free alternative for gene expression studies.
• The new algorithm enables the analysis of larger sample sizes, improving the applicability of exact statistical tests.
• This method facilitates more robust multiple testing procedures in high-throughput genomic data analysis.