Sample size calculation for differential expression analysis of RNA-seq data under Poisson distribution

Abstract

Sample size determination is an important issue in the experimental design of biomedical research. Because of the complexity of RNA-seq experiments, however, the field currently lacks a sample size method widely applicable to differential expression studies utilizing RNA-seq technology. In this report, we propose several methods for sample size calculation for single-gene differential expression analysis of RNA-seq data under Poisson distribution. These methods are then extended to multiple genes, with consideration for addressing the multiple testing problem by controlling false discovery rate. Moreover, most of the proposed methods allow for closedform sample size formulas with specification of the desired minimum fold change and minimum average read count, and thus are not computationally intensive. Simulation studies to evaluate the performance of the proposed sample size formulas are presented; the results indicate that our methods work well, with achievement of desired power. Finally, our sample size calculation methods are applied to three real RNA-seq data sets.

Inference and Sample Size Calculations Based on Statistical Tests in a Negative Binomial Distribution for Differential Gene Expression in RNAseq Data

“…The logarithmic transformation is usually adopted for skewness correction and variance stabilization as suggested by these studies [16,24]. The statistical inference on the quantity is ( )…”

“…The minimum of average read counts among these genes in the control group served as pilot data was estimated as u 0 =1 [16]. In addition, the sample sizes were estimated using u 0 =5,10 and 20.…”

“…Moreover, the Wald test based sample size calculation method is easier to compute and faster in an RNA-seq experimental design. Later, several sample size calculation methods that were derived from the score statistic and the log-likelihood ratio test (LRT) statistic using the Poisson distribution were proposed [16]. However, the assumption of a Poisson distribution that the expected mean and variance are equal usually does not hold for RNA-seq studies, where the variance is typically greater than the mean of the read counts [17].…”

“…For high dimensional microarray data analysis, an extension of the FDR correction was proposed [21] and is widely used by many researchers. To address similar issues for high-dimensional RNA-seq data, a sample size determination based on the extended FDR correction from microarray data analysis was further proposed [16].…”

“…Then, we derive sample size calculations based on these defined statistical tests. Lastly, we derived sample size calculation methods for testing multiple genes while controlling the FDR [16]. In Section 3, we simulated power for testing single gene and multiple genes corresponding to the sample sizes estimated from our proposed methods and an existing method.…”