A robust and fast two-sample test of equal correlations with an application to differential co-expression

Area of Science:

• Biostatistics
• Bioinformatics
• Genomics

Background:

• Accurate testing for equal Pearson correlation coefficients (PCCs) is crucial for biological analyses like differential co-expression.
• Existing methods often struggle with robustness to non-normal distributions, accuracy in small samples, and computational efficiency.
• There is a need for a reliable and fast method to test differential correlation.

Purpose of the Study:

• To introduce DICOSAR, a novel method for testing differential correlation.
• To provide a robust, accurate, and computationally efficient alternative to existing methods.
• To enable rapid testing of PCC equality in large-scale biological data analysis.

Main Methods:

• DICOSAR utilizes a saddlepoint approximation of the residual bootstrap.
• It integrates pooled residual bootstrap, signed root of a likelihood ratio statistic, and multivariate saddlepoint approximation.
• The method is designed for robustness, accuracy, and computational efficiency.

Main Results:

• DICOSAR demonstrates accurate control of type I error rates in differential correlation detection.
• It is robust against deviations from normal distributions and performs well with small sample sizes.
• DICOSAR proves to be a faster alternative to traditional bootstrap and permutation methods.
• The method is also applicable to testing differential correlation matrices.

Conclusions:

• DICOSAR offers a robust, accurate, and efficient analytical approach for testing the equality of PCCs.
• It facilitates rapid differential correlation testing, particularly in large-scale biological analyses.
• This method enhances the analysis of differential co-expression and related biological problems.