Differential enrichment/expression test

test_diff performs a differential enrichment/expression test based on protein/peptide-wise linear models and empirical Bayes statistics using limma. False Discovery Rates are estimated using "Strimmer's qvalue", "Benjamini-Hochberg fdr" or "Storey's qvalue".

Usage

test_diff(
  se,
  type = c("all", "control", "manual"),
  control = NULL,
  test = NULL,
  design_formula = formula(~0 + condition),
  fdr.type = c("Strimmer's qvalue(t)", "Strimmer's qvalue(p)", "BH", "Storey's qvalue")
)

Arguments

se: SummarizedExperiment, Proteomics data from quantity table (output from make_se(), make_se_parse()), or from re-aggregate peptide quantity from aggregateFeatures() ). For proteomics data directly from quantity table, it is adviced to first remove low quanlity and undesired rows by filter_se(), and missing values are imputed by impute(). Normalization is recommended in both workflow.
type: "control", "all" or "manual", The type of contrasts that will be tested. This can be all possible pairwise comparisons ("all"), limited to the comparisons versus the control ("control"), or manually defined contrasts ("manual").
control: Character(1), The condition to which contrasts are generated if type = "control" (a control condition would be most appropriate).
test: Character, The contrasts that will be tested if type = "manual". These should be formatted as "conditionA_vs_conditionB" or c("conditionA_vs_conditionC", "conditionB_vs_conditionC").
design_formula: Formula, Used to create the design matrix.
fdr.type: Character(1), the method to control false discovery rate, one of "Strimmer's qvalue(t)","Strimmer's qvalue(p)","BH","Storey's qvalue". "Strimmer's qvalue" calculate fdr vis fdrtool() using t-statistic or p values from limma. "BH" calculate fdr vis p.adjust() using "BH" method . "Storey's qvalue" calculate fdr vis stats::qvalue().

Value

A SummarizedExperiment object or a DEGdata containing diff, test result from limma and fdr estimates of differential expression.

Examples

# Load example
data(Silicosis_pg)
data <- Silicosis_pg
data_unique <- make_unique(data, "Gene.names", "Protein.IDs", delim = ";")

# Make SummarizedExperiment
ecols <- grep("LFQ.", colnames(data_unique))


## Load experiement design
data(Silicosis_ExpDesign)
exp_design <- Silicosis_ExpDesign
se <- make_se(data_unique, ecols, exp_design)

# Filter and normalize
filt <- filter_se(se, thr = 0, fraction = 0.4, filter_formula = ~ Reverse != "+" & Potential.contaminant!="+")
#> filter base on missing number is <= 0 in at least one condition.
#> filter base on missing number fraction < 0.4 in each row
#> filter base on giving formula 
norm <- normalize_vsn(filt)
#> vsn2: 8762 x 20 matrix (1 stratum). 
#> Please use 'meanSdPlot' to verify the fit.

# Impute missing values using different functions
imputed <- impute(norm, fun = "MinProb", q = 0.05)
#> Imputing along margin 2 (samples/columns).
#> [1] 0.3026531

# Test for differentially expressed proteins
diff <- test_diff(imputed, "control", "PBS")
#> Tested contrasts: W10_vs_PBS, W2_vs_PBS, W4_vs_PBS, W6_vs_PBS, W9_vs_PBS
#> Strimmer's qvalue(t)
diff <- test_diff(imputed, "manual", test = "PBS_vs_W6")
#> Tested contrasts: PBS_vs_W6
#> Strimmer's qvalue(t)
diff <- test_diff(imputed, "manual", test = "PBS_vs_W6", fdr.type = "Storey's qvalue")
#> Tested contrasts: PBS_vs_W6
#> Storey's qvalue