superdelta2: Main function of superdelta2 package to implement...

In fhlsjs/superdelta2: Superdelta procedure for multiple group differential gene expression analysis

Description

This function takes in the raw read count data matrix from RNA-Seq experiment, with rows corresponding to genes and columns corresponding to samples/libraries, and conducts differential expression analysis based on the robust superdelta2 method. Whether or not to use the total number of mapped read counts per-library as sample weights (the W parameter) is up to user's decision.

Usage

  superdelta2(mydata, offset = 1, Grps, W = NULL, trim = .2,
  adjp.thresh = 0.05, prop = 1.0)

Arguments

mydata	A data matrix with rows being genes and columns being samples. Note that this argument requires RAW read count of RNA-seq gene expression data, but not pre-processed data.
offset	A value to be added when taking logarithm to avoid log of zero. Default value is 1. Commonly used values include 0.5, 1, 2, and 3.
Grps	A character vector of length equal to the total number of samples in the data matrix, indicating group labels. This object is used to create design matrix for further computation.
W	Sample weights to be used. Default is NULL, which means unweighted (equally weighted). Users may want to use per-sample total read counts as the weighting metric. This is shown to have marginal gain in some cases.
trim	Trimming proportion when removing the most extreme values of between-group sum of squares. Default is 0.2.
adjp.thresh	Cutoff of adjusted p-value to define significance. Default is 0.05.
prop	The proportion of reference genes used in the main (second round) spherical trimming for bias correction. The default value is 1.0, which means using all genes for bias correction. Using a small proportion can improve the computational speed, at the cost of slightly less accurate bias correction.

Details

If not NULL, W must be a vector of positive per-sample weights of length equal to the number of columns (samples) in the data matrix, to specify per-sample weights. The program will automatically normalize the weights so that they sum up to 1. We recommend using the total number of mapped read counts as per-sample weights.

Value

The superdelta2 function will return a list with the following objects:

SigGenes	A named vector of the significant gene IDs. Typically the name of this vector is inherited from the rownames of the input data matrix.
Fstats	A vector of length equal to the number of genes in the data matrix indicating the super-delta F-statistic for each gene.
pvalues.F	Raw p-values obtained from the super-delta F-statistics.
padj.F	Benjamini adjusted p-values obtained from the super-delta F-statistics.
Log2FC	Log2 fold-change of post-hoc pairwise t-tests. Technically this is equal to the numerator of Tukey style t-statistics.
tstats	A matrix of row dimension equal to the number of genes in the input data matrix, providing the super-delta post-hoc Tukey t-statistics.
pvalues.t	Raw p-values obtained from the super-delta Tukey t-statistics.
padj.t	Benjamini adjusted p-values obtained from the super-delta Tukey t-statistics.
WBGRMS	Estimated weighted between-group mean square, namely the top part of F-statistics.
WWGRMS	Estimated weighted within-group mean square, namely the bottom part of F-statistics.
sigma2hat	Estimated sigma square (variance of the error terms).
Rtrim	Saved R object containing the information of the main trim.

Note

The superdelta2 function implements a general-purpose differential gene expression analysis procedure that can be used in a multiple group (>=3) setting. It inherits the basic idea of superdelta (see References for details) with a robust internal normalization and an asymptotically unbiased estimator of the “oracle” between-group difference. It applies a robust trimming procedure to the estimated between-group sum of squares and takes advantage of all the within-group sum of squares, to end up with the superdelta F-statistics. In addition, superdelta2 is accompanied with a Tukey style pairwise comparison to also obtain post-hoc t-statistics.

Author(s)

Yuhang Liu, Xing Qiu, Jinfeng Zhang, and Zihan Cui

References

Liu, Y., Zhang, J., & Qiu, X. (2017). Super-delta: a new differential gene expression analysis procedure with robust data normalization. BMC bioinformatics, 18(1), 582.

See Also

SIM1, SIM2, SIM3

Examples

  ## Load the sample data
  data(SampleData)
  ## Number of genes and samples
  ngenes <- 5000; n1 <- n2 <- n3 <- 50; ns <- c(n1,n2,n3)
  Groups <- c(rep("A",n1), rep("B",n2), rep("C",n3))
  mod1 <- superdelta2(mydata = SIM1, offset = 1, Grps = Groups)
  mod2 <- superdelta2(mydata = SIM2, offset = 1, Grps = Groups)
  mod3 <- superdelta2(mydata = SIM3, offset = 1, Grps = Groups)

fhlsjs/superdelta2 documentation built on Sept. 15, 2020, 12:03 a.m.