Compute q-values

Source: R/compute_metrics.R

This function computes q-values from the posterior error probabilities (PEPs). The functions takes the PEPs from the given assay's rowData and adds a new variable to it that contains the computed q-values.

pep2qvalue(object, i, groupBy, PEP, rowDataName = "qvalue")

Arguments

object

A QFeatures object

i

A numeric() or character() vector indicating from which assays the rowData should be taken.

groupBy

A character(1) indicating the variable name in the rowData that contains the grouping variable, for instance to compute protein FDR. When groupBy is not missing, the best feature approach is used to compute the PEP per group, meaning that the smallest PEP is taken as the PEP of the group.

PEP

A character(1) indicating the variable names in the rowData that contains the PEPs. Since, PEPs are probabilities, the variable must be contained in (0, 1).

rowDataName

A character(1) giving the name of the new variable in the rowData where the computed FDRs will be stored. The name cannot already exist in any of the assay rowData.

Value

A QFeatures object.

Details

The q-value of a feature (PSM, peptide, protein) is the minimum FDR at which that feature will be selected upon filtering (Savitski et al.). On the other hand, the feature PEP is the probability that the feature is wrongly matched and hence can be seen as a local FDR (Kall et al.). While filtering on PEP is guaranteed to control for FDR, it is usually too conservative. Therefore, we provide this function to convert PEP to q-values.

We compute the q-value of a feature as the average of the PEPs associated to PSMs that have equal or greater identification confidence (so smaller PEP). See Kall et al. for a visual interpretation.

We also allow inference of q-values at higher level, for instance computing the protein q-values from PSM PEP. This can be performed by supplying the groupBy argument. In this case, we adopt the best feature strategy that will take the best (smallest) PEP for each group (Savitski et al.).

References

Käll, Lukas, John D. Storey, Michael J. MacCoss, and William Stafford Noble. 2008. “Posterior Error Probabilities and False Discovery Rates: Two Sides of the Same Coin.” Journal of Proteome Research 7 (1): 40–44.

Savitski, Mikhail M., Mathias Wilhelm, Hannes Hahne, Bernhard Kuster, and Marcus Bantscheff. 2015. “A Scalable Approach for Protein False Discovery Rate Estimation in Large Proteomic Data Sets.” Molecular & Cellular Proteomics: MCP 14 (9): 2394–2404.

Examples

data("scp1")
scp1 <- pep2qvalue(scp1,
                   i = 1,
                   groupBy = "protein",
                   PEP = "dart_PEP",
                   rowDataName = "qvalue_protein")
## Check results
rowData(scp1)[[1]][, c("dart_PEP", "qvalue_protein")]
#> DataFrame with 166 rows and 2 columns
#>              dart_PEP qvalue_protein
#>             <numeric>      <numeric>
#> PSM3773   7.78683e-05    1.24198e-16
#> PSM9078   3.51840e-01    7.55061e-02
#> PSM9858   1.54857e-04    7.83117e-07
#> PSM11744  1.00000e+00    2.28967e-01
#> PSM21752  3.42850e-01    5.52325e-02
#> ...               ...            ...
#> PSM732069 4.89848e-05    4.41181e-09
#> PSM735396 3.89829e-07    1.24198e-16
#> PSM744756 1.54479e-06    7.82514e-08
#> PSM745037 4.44010e-01    1.14849e-01
#> PSM745130 2.07957e-04    1.83409e-10