About this vignette

This vignette is dedicated to advanced users and to method developers. It assumes that you are already familiar with QFeatures and scp and that you are looking for more flexibility in the analysis of your single-cell proteomics (SCP) data. In fact, scp provides wrapper functions around generic functions and metrics. However, advanced users may want to apply or develop their own features. The QFeatures class offers a flexible data container while guaranteeing data consistency.

In this vignette, you will learn how to:

  • Modify the quantitative data
  • Modify the sample metadata
  • Modify the feature metadata
  • Create a new function for scp

As a general guideline, you can add/remove/update data in a QFeatures in 4 main steps:

  1. Gather the data to change and other required data involved in the processing.
  2. Apply the transformation/computation.
  3. Insert the changes in the QFeatures object.
  4. Make sure the updated QFeatures object is still valid.

To illustrate the different topics, we will load the scp1 example data.

#> An instance of class QFeatures containing 5 assays:
#>  [1] 190321S_LCA10_X_FP97AG: SingleCellExperiment with 166 rows and 11 columns 
#>  [2] 190222S_LCA9_X_FP94BM: SingleCellExperiment with 176 rows and 11 columns 
#>  [3] 190914S_LCB3_X_16plex_Set_21: SingleCellExperiment with 215 rows and 16 columns 
#>  [4] peptides: SingleCellExperiment with 539 rows and 38 columns 
#>  [5] proteins: SingleCellExperiment with 292 rows and 38 columns

Modify the quantitative data

To illustrate how to modify quantitative data, we will implement a normByType function that will normalize the feature (row) for each cell type separately. This function is probably not relevant for a real case analysis, but it provides a good example of a custom data processing. The process presented in this section is applicable to any custom function that takes at least a matrix-like object as input and returns a matrix-like object as output.

normByType <- function(x, type) {
    ## Check argument
    stopifnot(length(type) == ncol(x))
    ## Normalize for each type separately
    for (i in type) {
        ## Get normalization factor
        nf <- rowMedians(x[, type == i], na.rm = TRUE)
        ## Perform normalization
        x[, type == i] <- x[, type == i] / nf
    ## Return normalized data

Suppose we want to apply the function to the proteins assay, we need to first extract that assay. We here need to transfer the sample metadata from the QFeatures object to the extracted SingleCellExperiment in order to get the sample types required by the normByType function. We therefore use getWithColData.

sce <- getWithColData(scp1, "proteins")
#> Warning: 'experiments' dropped; see 'metadata'
#> class: SingleCellExperiment 
#> dim: 292 38 
#> metadata(0):
#> assays(2): assay aggcounts
#> rownames(292): A1A519 A5D8V6 ... REV__CON__Q32PI4 REV__CON__Q3MHN5
#> rowData names(9): protein Match.time.difference ...
#>   Potential.contaminant .n
#> colnames(38): 190321S_LCA10_X_FP97AG_RI1 190321S_LCA10_X_FP97AG_RI2 ...
#>   190914S_LCB3_X_16plex_Set_21_RI15 190914S_LCB3_X_16plex_Set_21_RI16
#> colData names(7): Set Channel ... sortday digest
#> reducedDimNames(0):
#> mainExpName: NULL
#> altExpNames(0):

Next, we can apply the data transformation to the quantitative data. As mentioned above, our function expects a matrix-like object as an input, so we use the assay function. We then update the SingleCellExperiment object.

mnorm <- normByType(assay(sce), type = sce$SampleType)
assay(sce) <- mnorm

We are now faced with 2 possibilities: either we want to create a new assay or we want to overwrite an existing assay. In both cases we need to make sure your data is still valid after data transformation.

Create a new assay

Creating a new assay has the advantage that you don’t modify an existing assay and hence limit the risk of introducing inconsistency in the data and avoid losing intermediate steps of the data processing.

We add the transformed assay using the addAssay function, then link the parent assay to the transformed assay using addAssayLinkOneToOne. Note that if each row name in the parent assay does not match exactly one row in the child assay, you can also use addAssayLink that will require a linking variable in the rowData.

scp1 <- addAssay(scp1, sce, name = "proteinsNorm")
scp1 <- addAssayLinkOneToOne(scp1, from = "proteins", to = "proteinsNorm")
#> An instance of class QFeatures containing 6 assays:
#>  [1] 190321S_LCA10_X_FP97AG: SingleCellExperiment with 166 rows and 11 columns 
#>  [2] 190222S_LCA9_X_FP94BM: SingleCellExperiment with 176 rows and 11 columns 
#>  [3] 190914S_LCB3_X_16plex_Set_21: SingleCellExperiment with 215 rows and 16 columns 
#>  [4] peptides: SingleCellExperiment with 539 rows and 38 columns 
#>  [5] proteins: SingleCellExperiment with 292 rows and 38 columns 
#>  [6] proteinsNorm: SingleCellExperiment with 292 rows and 38 columns

Overwrite an existing assay

Overwriting an existing assay has the advantage to limit the memory consumption as compared to adding a new assay. You can overwrite an assay simply by replacing the target assay in its corresponding slot.

scp1[["proteins"]] <- sce

Check for validity

Applying custom changes to the data increases the risk for data inconsistencies. You can however verify that a QFeatures object is still valid after some processing thanks to the validObject function.

#> [1] TRUE

If the function detects no issues in the data, it will return TRUE. Otherwise the function will throw an informative error that should guide the user to identifying the issue.

Modify the sample metadata

To illustrate how to modify the sample metadata, we will compute the average expression in each sample and include to the colData of the QFeatures object. This is typically performed when computing QC metrics for sample filtering. So, we first extract the required data, in this case the quantitative values, and compute the sample-wise average protein expression.

m <- assay(scp1, "proteins")
meanExprs <- colMeans(m, na.rm = TRUE)
#>        190321S_LCA10_X_FP97AG_RI1        190321S_LCA10_X_FP97AG_RI2 
#>                        45368.3628                         7566.8782 
#>        190321S_LCA10_X_FP97AG_RI3        190321S_LCA10_X_FP97AG_RI4 
#>                         3271.0402                         2158.4230 
#>        190321S_LCA10_X_FP97AG_RI5        190321S_LCA10_X_FP97AG_RI6 
#>                         1162.3936                         1835.7939 
#>        190321S_LCA10_X_FP97AG_RI7        190321S_LCA10_X_FP97AG_RI8 
#>                         1293.4738                         1415.4205 
#>        190321S_LCA10_X_FP97AG_RI9       190321S_LCA10_X_FP97AG_RI10 
#>                         1965.6531                         1264.4064 
#>       190321S_LCA10_X_FP97AG_RI11         190222S_LCA9_X_FP94BM_RI1 
#>                         1238.5979                        48901.4654 
#>         190222S_LCA9_X_FP94BM_RI2         190222S_LCA9_X_FP94BM_RI3 
#>                         5189.5912                         2802.1850 
#>         190222S_LCA9_X_FP94BM_RI4         190222S_LCA9_X_FP94BM_RI5 
#>                         1204.2350                          715.7667 
#>         190222S_LCA9_X_FP94BM_RI6         190222S_LCA9_X_FP94BM_RI7 
#>                         1032.7122                          403.8267 
#>         190222S_LCA9_X_FP94BM_RI8         190222S_LCA9_X_FP94BM_RI9 
#>                         1089.0548                         1076.2613 
#>        190222S_LCA9_X_FP94BM_RI10        190222S_LCA9_X_FP94BM_RI11 
#>                         1572.3337                         1186.4431 
#>  190914S_LCB3_X_16plex_Set_21_RI1  190914S_LCB3_X_16plex_Set_21_RI2 
#>                        96881.8350                         2731.9430 
#>  190914S_LCB3_X_16plex_Set_21_RI3  190914S_LCB3_X_16plex_Set_21_RI4 
#>                         8208.1958                          208.2857 
#>  190914S_LCB3_X_16plex_Set_21_RI5  190914S_LCB3_X_16plex_Set_21_RI6 
#>                         3209.7730                         2014.4889 
#>  190914S_LCB3_X_16plex_Set_21_RI7  190914S_LCB3_X_16plex_Set_21_RI8 
#>                         2897.1103                         3006.6741 
#>  190914S_LCB3_X_16plex_Set_21_RI9 190914S_LCB3_X_16plex_Set_21_RI10 
#>                         2760.8564                         2549.9348 
#> 190914S_LCB3_X_16plex_Set_21_RI11 190914S_LCB3_X_16plex_Set_21_RI12 
#>                         2388.5289                         2533.8006 
#> 190914S_LCB3_X_16plex_Set_21_RI13 190914S_LCB3_X_16plex_Set_21_RI14 
#>                         2134.5615                         3125.1980 
#> 190914S_LCB3_X_16plex_Set_21_RI15 190914S_LCB3_X_16plex_Set_21_RI16 
#>                         3276.5618                         3195.8827

Next, we insert the computed averages into the colData. You need to make sure to match sample names because an extracted assay may not contain all samples and may be in a different order compared to the colData.

colData(scp1)[names(meanExprs), "meanProtExprs"] <- meanExprs

The new sample variable meanProtExprs is now accessible for filtering or plotting. The $ operator makes it straightforward to access the new data.


To make sure that the process did not corrupt the colData, we advise to verify the data is still valid.

#> [1] TRUE

Modify the feature metadata

We will again illustrate how to modify the feature metadata with an example. We here demonstrate how to add the number of samples in which each feature is detected for the three first assays (PSM assays). The challenge here is that the metric needs to be computed for each assay separately and inserted in the corresponding assay.

We will take advantage of the replacement function for rowData as implemented in QFeatures. It expects a list-like object where names indicate in which assays we want to modify the rowData and each element contains a table with the replacement values.

We therefore compute the metrics for each assay separately and store the results in a named List.

## Initialize the List object that will store the computed values
res <- List()
## We compute the metric for the first 3 assays
for (i in 1:3) {
    ## We get the quantitative values for the current assay
    m <- assay(scp1[[i]])
    ## We compute the number of samples in which each features is detected
    n <- rowSums(!is.na(m) & m != 0)
    ## We store the result as a DataFrame in the List
    res[[i]] <- DataFrame(nbSamples = n)
names(res) <- names(scp1)[1:3]
#> List of length 3
#> names(3): 190321S_LCA10_X_FP97AG 190222S_LCA9_X_FP94BM 190914S_LCB3_X_16plex_Set_21
#> DataFrame with 166 rows and 1 column
#>           nbSamples
#>           <numeric>
#> PSM3773          11
#> PSM9078          11
#> PSM9858          11
#> PSM11744         11
#> PSM21752          0
#> ...             ...
#> PSM732069        11
#> PSM735396        11
#> PSM744756        10
#> PSM745037        11
#> PSM745130        11

Now that we have a List of DataFrames, we can update the object.

rowData(scp1) <- res

The new feature variable nbSamples is now accessible for filtering or plotting. The rbindRowData function facilitates the access the new data.

rd <- rbindRowData(scp1, i = 1:3)
ggplot(data.frame(rd)) +
    aes(x = nbSamples) +
    geom_histogram(bins = 16) +
    facet_wrap(~ assay)

To make sure that the process did not corrupt the rowData in any assay, we advise to verify the data is still valid.

#> [1] TRUE

Create a new function for scp

The modifying data in a QFeatures involves a multiple-step process. Creating a wrapper function that would take care of those different steps in a single line of code is a good habit to reduce the length of analysis scripts and hence making it easier to understand and less error-prone.

We will wrap the last example in a new function that we call computeNbDetectedSamples.

computeNbDetectedSamples <- function(object, i) {
    res <- List()
    for (ii in i) {
        m <- assay(object[[ii]])
        n <- rowSums(!is.na(m) & m != 0)
        res[[ii]] <- DataFrame(nbSamples = n)
    names(res) <- names(object)[i]
    rowData(object) <- res

Thanks to this new function, the previous section now simply boils down to running:

scp1 <- computeNbDetectedSamples(scp1, i = 1:3)

Keep in mind a few recommendations when creating a new function for scp:

  • The function should take a QFeatures object as input (and more arguments if needed) and return a QFeatures object as output. This will make workflows much easier to understand.
  • Allow user to select assays (if required) either as numeric, character, or logical.
  • Use conventional argument names: when naming an argument, try to match the names that already exist. For instance, selecting assays is passed through the i argument, selecting rowData variables is passed through rowvars and selecting colData variables is passed through colvars.
  • Follow the rformassspectrometry coding style

What’s next?

So you developed a new metric or method and believe it might benefit the community? We would love to hear about your improvements and eventually include your new functionality into scp or associate your new package to our documentation. Please, raise an issue in our Github repository to suggest your improvements or, better, submit your code as a pull request.

Session information

R version 4.1.1 (2021-08-10)
Platform: x86_64-pc-linux-gnu (64-bit)
Running under: Ubuntu 20.04.3 LTS

Matrix products: default
BLAS/LAPACK: /usr/lib/x86_64-linux-gnu/openblas-pthread/libopenblasp-r0.3.8.so

 [1] LC_CTYPE=en_US.UTF-8       LC_NUMERIC=C              
 [3] LC_TIME=en_US.UTF-8        LC_COLLATE=en_US.UTF-8    
 [5] LC_MONETARY=en_US.UTF-8    LC_MESSAGES=C             
 [7] LC_PAPER=en_US.UTF-8       LC_NAME=C                 
 [9] LC_ADDRESS=C               LC_TELEPHONE=C            

attached base packages:
[1] stats4    stats     graphics  grDevices utils     datasets  methods  
[8] base     

other attached packages:
 [1] ggplot2_3.3.5                scp_1.3.3                   
 [3] QFeatures_1.3.7              MultiAssayExperiment_1.19.16
 [5] SummarizedExperiment_1.23.5  Biobase_2.53.0              
 [7] GenomicRanges_1.45.0         GenomeInfoDb_1.29.8         
 [9] IRanges_2.27.2               S4Vectors_0.31.5            
[11] BiocGenerics_0.39.2          MatrixGenerics_1.5.4        
[13] matrixStats_0.61.0           BiocStyle_2.21.3            

loaded via a namespace (and not attached):
 [1] sass_0.4.0                  jsonlite_1.7.2             
 [3] bslib_0.3.1                 BiocManager_1.30.16        
 [5] highr_0.9                   GenomeInfoDbData_1.2.7     
 [7] yaml_2.2.1                  pillar_1.6.3               
 [9] lattice_0.20-45             glue_1.4.2                 
[11] digest_0.6.28               XVector_0.33.0             
[13] colorspace_2.0-2            htmltools_0.5.2            
[15] Matrix_1.3-4                pkgconfig_2.0.3            
[17] bookdown_0.24               zlibbioc_1.39.0            
[19] purrr_0.3.4                 scales_1.1.1               
[21] tibble_3.1.5                farver_2.1.0               
[23] generics_0.1.0              AnnotationFilter_1.17.1    
[25] ellipsis_0.3.2              withr_2.4.2                
[27] cachem_1.0.6                lazyeval_0.2.2             
[29] cli_3.0.1                   magrittr_2.0.1             
[31] crayon_1.4.1                memoise_2.0.0              
[33] evaluate_0.14               fs_1.5.0                   
[35] fansi_0.5.0                 MASS_7.3-54                
[37] textshaping_0.3.5           tools_4.1.1                
[39] lifecycle_1.0.1             stringr_1.4.0              
[41] munsell_0.5.0               cluster_2.1.2              
[43] DelayedArray_0.19.4         compiler_4.1.1             
[45] pkgdown_1.9000.9000.9000    jquerylib_0.1.4            
[47] systemfonts_1.0.2           rlang_0.99.0.9000          
[49] grid_4.1.1                  RCurl_1.98-1.5             
[51] MsCoreUtils_1.5.0           SingleCellExperiment_1.15.2
[53] igraph_1.2.6                labeling_0.4.2             
[55] bitops_1.0-7                rmarkdown_2.11             
[57] gtable_0.3.0                R6_2.5.1                   
[59] knitr_1.36                  dplyr_1.0.7                
[61] fastmap_1.1.0               utf8_1.2.2                 
[63] clue_0.3-60                 rprojroot_2.0.2            
[65] ragg_1.1.3                  ProtGenerics_1.25.1        
[67] desc_1.4.0                  stringi_1.7.5              
[69] vctrs_0.3.8                 tidyselect_1.1.1           
[71] xfun_0.26