DRomics (version 2.6-2)

itemselect: Selection of significantly responsive items

Description

Significantly responsive items are selected using one of the three proposed methods: a quadratic trend test, a linear trend test or an ANOVA-based test.

Usage

itemselect(omicdata, select.method = c("quadratic", "linear", "ANOVA"), 
  FDR = 0.05, max.ties.prop = 0.2)   
# S3 method for itemselect
print(x, nfirstitems = 20, ...)

Value

itemselect returns an object of class "itemselect", a list with 5 components:

adjpvalue

the vector of the p-values adjusted by the Benjamini-Hochberg method (also called q-values) for selected items (adjpvalue inferior to FDR) sorted in ascending order

selectindex

the corresponding vector of row indices of selected items in the object omicdata

omicdata

The corresponding object of class "microarraydata", "RNAseqdata", "continuousomicdata" or "continuousanchoringdata" given in input.

select.method

The selection method given in input.

FDR

The threshold in term of FDR given in input.

The print of a "itemselect" object gives the number of selected items and the identifiers of the 20 most responsive items.

Arguments

omicdata

An object of class "microarraydata", "RNAseqdata", "metabolomicdata" or "continuousanchoringdata" respectively returned by functions microarraydata, RNAseqdata, metabolomicdata or continuousanchoringdata.

select.method

"quadratic" for a quadratic trend test on dose ranks, "linear" for a linear trend test on dose ranks and "ANOVA" for an ANOVA-type test (see details for further explaination).

FDR

The threshold in term of FDR (False Discovery Rate) for selecting responsive items.

max.ties.prop

The maximal tolerated proportion of tied values for each item, above which the item cannot be selected (must be in ]0, 0.5], and by default fixed at 0.2 - see details for a description of this filtering step).

x

An object of class "itemselect".

nfirstitems

The maximum number of selected items to print.

...

further arguments passed to print function.

Author

Marie-Laure Delignette-Muller

Details

The selection of responsive items is performed using the limma package for microarray and continuous omics data (such as metabolomics), the DESeq2 package for RNAseq data and the lm function for continuous anchoring data. Three methods are proposed (as described below). Within limma those methods are implemented using functions lmFit, eBayes and topTable with p-values ajusted for multiple testing using the Benjamini-Hochberg method (also called q-values), with the false discovery rate given in input (argument FDR). Within DESeq2 those methods are implemented using functions DESeqDataSetFromMatrix, DESeq and results with p-values ajusted for multiple testing using the Benjamini-Hochberg method (also called q-values), with the false discovery rate given in input (argument FDR). For continuous anchoring data, the lm and anova functions are used to fit the model and compare it to the null model, and the pvalues are then corrected using the function p.adjust with the Benjamini-Hochberg method.

• The ANOVA_based test ("ANOVA") is classically used for selection of omics data in the general case but it requires many replicates per dose to be efficient, and is thus not really suited for a dose-response design.

• The linear trend test ("linear") aims at detecting monotonic trends from dose-response designs, whatever the number of replicates per dose. As proposed by Tukey (1985), it tests the global significance of a linear model describing the response as a function of the dose in rank-scale.

• The quadratic trend test ("quadratic") tests the global significance of a quadratic model describing the response as a function of the dose in rank-scale. It is a variant of the linear trend method that aims at detecting monotonic and non monotonic trends from a dose-response designs, whatever the number of replicates per dose (default chosen method).

After the use of one this previously described tests, a filter based on the proportion of tied values is also performed whatever the type of data, assuming tied values correspond to a minimal common value at which non detections were imputed. All items having a proportion of such tied minimal values above the input argument max.ties.prop are eliminated from the selection.

References

Tukey JW, Ciminera JL and Heyse JF (1985), Testing the statistical certainty of a response to increasing doses of a drug. Biometrics, 295-301.

Ritchie ME, Phipson B, Wu D, Hu Y, Law CW, Shi W, and Smyth, GK (2015), limma powers differential expression analyses for RNA-sequencing and microarray studies. Nucleic Acids Research 43, e47.

Love MI, Huber W, and Anders S (2014), Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2. Genome biology, 15(12), 550.

See Also

See lmFit, eBayes and topTable for details about the used functions of the limma package and DESeqDataSetFromMatrix, DESeq and results for details about the used functions of the DESeq2 package.

Examples

# (1) an example on a microarray data set (a subsample of a greater data set) 
#     
datafilename <- system.file("extdata", "transcripto_sample.txt", package="DRomics")

(o <- microarraydata(datafilename, check = TRUE, norm.method = "cyclicloess"))

# 1.a using the quadratic trend test
#
(s_quad <- itemselect(o, select.method = "quadratic", FDR = 0.05))
print(s_quad, nfirstitems = 30)

# to get the names of all the selected items
(selecteditems <- s_quad$omicdata$item[s_quad$selectindex]) 

# \donttest{

# 1.b using the linear trend test
#
(s_lin <- itemselect(o, select.method = "linear", FDR = 0.05))

# 1.c using the ANOVA-based test
#
(s_ANOVA <- itemselect(o, select.method = "ANOVA", FDR = 0.05))

# 1.d using the quadratic trend test with a smaller false discovery rate
#
(s_quad.2 <- itemselect(o, select.method = "quadratic", FDR = 0.001))

# }