Methods Formula with relevant details FDR computation Note
T-test


where is the average level of protein i in sample group j;s(i) is the pooled standard error:


sj2(i) is the variance of protein i in sample group j and  nj denotes the number of samples in group j.

 Benjamini-Hochberg method Assumes normal distribution
SAM
where and s(i) are the same as in T-test. s0 >0 is a constant added to avoid the dependency of d(i) on the protein abundance level.		 permutation-based approach  
LIMMA
where is the same as in T-test,  is the posterior residual variance, a weighted average of prior and residual variance; u(i) is the unscaled standard deviation.		 Benjamini-Hochberg method A linear model to determine differential expression
Wilcoxon rank sum test
where
   and ni is sample size for group i, T1 is sum of ranks for group 1.		 Benjamini-Hochberg method Non parametric, normal distribution approximation
Rank Product   ,
where rlup(i) (or rldown(i), respectively) are the ranks of protein i in the ordered protein lists of length kl sorted in decreasing or increasing order for replicate l; n is the total number of replicates.		 permutation-based approach Non parametric
ROTS (maximize the reproducibility of detections in a family of modified t-statistics)
where and s(i) are the same as in T-test. The parameters α0 and α1 are determined so that they maximize the reproducibility Z- score
Zk,a = (Rk,a- R0k,a) / sk,a ,
where k is the top list sizes; α = (α0, α1);Rk,α and R0k,α are the observed and random reproducibility; sk,α is the standard deviation of the bootstrap distribution. Reproducibility is the overlap of the top-ranked proteins across bootstrap re-samples.		 permutation-based approach Special cases:  the T-test (α0 = 0, α1 = 1), the signal log-ratio (α0 = 1, α1 = 0), the SAM statistic (α1 = 1, α0 is a percentile of the standard deviation).
Table 1: Summary of six statistical tools used to determine differential expression: T-test, SAM, LIMMA, Wilcoxon rank sum test, Rank Product and ROTS.